Remove redundant prepared input APIs

.gitignore

@@ -2,6 +2,8 @@
 .codex
 AGENTS.md
 
+.DS_Store
+
 # ---> Go
 # If you prefer the allow list template instead of the deny list, see community template:
 # https://github.com/github/gitignore/blob/main/community/Golang/Go.AllowList.gitignore
@@ -22,6 +24,9 @@ AGENTS.md
 # Dependency directories (remove the comment below to include it)
 # vendor/
 
+# Go cache
+.gocache
+
 # Go workspace file
 go.work
 go.work.sum

.woodpecker/release.yml

@@ -2,8 +2,33 @@ when:
   - event: tag
 
 steps:
-  - name: build-release-assets
+  validate:
     image: golang:1.25
+    commands:
+      - go test ./...
+      - go test -race ./...
+      - go vet ./...
+      - go build ./...
+      - go test ./internal/doccheck
+      - go test ./internal/config -run '^TestExamplesLoadAndValidate$'
+
+  cross-build:
+    image: golang:1.25
+    depends_on: validate
+    commands:
+      - |
+        set -eu
+        output_dir="$(mktemp -d)"
+        trap 'rm -rf "$output_dir"' EXIT
+        for target in linux/amd64 linux/arm64 darwin/amd64 darwin/arm64 windows/amd64 windows/arm64; do
+          goos="${target%/*}"
+          goarch="${target#*/}"
+          CGO_ENABLED=0 GOOS="$goos" GOARCH="$goarch" go build -o "$output_dir/narratio-$goos-$goarch" ./cmd/narratio
+        done
+
+  build-release-assets:
+    image: golang:1.25
+    depends_on: [validate, cross-build]
     commands:
       - |
         set -eu
@@ -33,7 +58,7 @@ steps:
         build_binary windows amd64 ".exe"
         build_binary windows arm64 ".exe"
 
-  - name: publish-release
+  publish-release:
     image: woodpeckerci/plugin-release
     depends_on:
       - build-release-assets

.woodpecker/shuffle.yml

@@ -0,0 +1,8 @@
+when:
+  - event: cron
+
+steps:
+  shuffled-race-tests:
+    image: golang:1.25
+    commands:
+      - go test -race -shuffle=on -count=3 ./...

.woodpecker/verify.yml

@@ -0,0 +1,47 @@
+when:
+  - event: [push, pull_request]
+
+steps:
+  tests:
+    image: golang:1.25
+    commands:
+      - go test ./...
+
+  race-tests:
+    image: golang:1.25
+    depends_on: tests
+    commands:
+      - go test -race ./...
+
+  static-analysis:
+    image: golang:1.25
+    depends_on: tests
+    commands:
+      - go vet ./...
+
+  build:
+    image: golang:1.25
+    depends_on: tests
+    commands:
+      - go build ./...
+
+  documentation-and-examples:
+    image: golang:1.25
+    depends_on: tests
+    commands:
+      - go test ./internal/doccheck
+      - go test ./internal/config -run '^TestExamplesLoadAndValidate$'
+
+  cross-build:
+    image: golang:1.25
+    depends_on: [race-tests, static-analysis, build, documentation-and-examples]
+    commands:
+      - |
+        set -eu
+        output_dir="$(mktemp -d)"
+        trap 'rm -rf "$output_dir"' EXIT
+        for target in linux/amd64 linux/arm64 darwin/amd64 darwin/arm64 windows/amd64 windows/arm64; do
+          goos="${target%/*}"
+          goarch="${target#*/}"
+          CGO_ENABLED=0 GOOS="$goos" GOARCH="$goarch" go build -o "$output_dir/narratio-$goos-$goarch" ./cmd/narratio
+        done

README.md

@@ -1,327 +1,42 @@
 # narratio
 
-`narratio` is a Go orchestration application for processing D&D session audio into transcripts and generated artifacts.
+Narratio is a stage-driven Go orchestrator for turning D&D session audio into
+polished transcripts, validated Notarius extraction lanes, and generated
+artifacts.
 
-## Current Implementation
+It runs a deterministic workflow with manifest-driven continuation, remote
+publish, and restore support.
 
-Implemented now:
-
-- strict config loading/validation (`pipeline.yml` and `session.yml`)
-- local workspace/session layout, locking, and manifest persistence
-- resumable stage control (`run`, `plan`, `resume`, `run-stage`, `status`)
-- real `prepare`, `transcribe`, `merge`, `polish`, `normalize`, `trim`, and `analyze` stages
-- real WhisperX, Seriatim, and Audita adapters
-- real Scriptorium subprocess adapter
-- optional Scriptorium render diagnostics (`render_debug`)
-
-Not implemented yet:
-
-- `archive` stage behavior
-- `notify` stage behavior
-- additional analyze artifacts beyond `session_recap`
-- generic DAG orchestration
-
-## Config Files
-
-Narratio expects two YAML files:
-
-- `pipeline.yml`: pipeline/workspace settings
-- `session.yml`: per-session settings
-
-Pipeline config lookup for CLI commands:
-
-- if `--config <path>` is provided, Narratio uses that path
-- if `--config` is omitted, Narratio searches in this order:
-  - `/usr/local/etc/narratio/pipeline.yml`
-  - `/etc/narratio/pipeline.yml`
-
-Optional secrets-from-files config:
-
-- `pipeline.secrets.env_dir` may point to a directory of secret files
-- each top-level file with an env-var-style name is loaded as an environment variable:
-  - file name = env var name
-  - file contents = env var value (trailing newline/CRLF trimmed)
-- process environment wins: existing env vars are not overwritten
-- if configured, Narratio fails fast when `env_dir` is missing/unreadable
-- relative `env_dir` values resolve from Narratio’s current working directory
-
-YAML decoding is strict (`KnownFields(true)`), so unknown fields fail fast.
-
-## Canonical Stage Order
-
-1. `prepare`
-2. `transcribe`
-3. `merge`
-4. `polish`
-5. `normalize`
-6. `trim`
-7. `analyze`
-8. `archive`
-9. `notify`
-
-## Transcript Tiers
-
-- `transcripts/merged.json`: canonical deterministic merged transcript from Seriatim merge
-- `transcripts/processed.json`: full raw Audita-polished transcript output
-- `transcripts/normalized.json`: Seriatim-normalized transcript from the normalize stage
-- `transcripts/trimmed.json`: gameplay-only normalized polished transcript from trim stage
-
-## Audita Configuration
-
-`pipeline.audita` configures the real Audita subprocess adapter used by `polish`.
-
-Required:
-
-- `binary`
-- `timeout`
-- `base_url`
-- `model`
-
-Optional:
-
-- `llm_api_key_env` (when set, Narratio requires that env var and passes it to Audita as `AUDITA_LLM_API_KEY`)
-- `modules` override list (when empty/omitted, Narratio does not pass `--modules`)
-- `transcript_description`
-- `config_path`
-- `output_schema` (`bare-segments` or `audita-v1`)
-- `work_dir_retention` (`always`, `auto`, or `never`)
-- `total_llm_concurrency` (> 0 when provided)
-- `proposal_llm_concurrency` (> 0 when provided)
-- `validation_model`
-- `validation_llm_concurrency` (> 0 when provided)
-- `report` (defaults to `true`)
-
-Narratio passes only configured optional Audita flags. Omitted optional values are left to Audita runtime defaults.
-
-## Normalize Configuration
-
-`pipeline.normalize` is optional. When omitted, Narratio defaults to:
-
-- `output_path: transcripts/normalized.json`
-- `output_schema: seriatim-intermediate`
-- `report: true`
-
-Allowed `normalize.output_schema` values:
-
-- `seriatim-minimal`
-- `seriatim-intermediate`
-- `seriatim-full`
-
-`normalize.output_path` is treated as session-workdir-relative when not absolute.
-
-Normalize stage behavior summary:
-
-- normalize runs after `polish` and before `trim`
-- normalize resolves `transcripts/processed.json`
-- normalize runs Seriatim `normalize` to produce `transcripts/normalized.json`
-- normalize diagnostics are written to:
-  - `artifacts/seriatim.normalize.report.json` (when enabled)
-  - `logs/seriatim.normalize.stdout.log`
-  - `logs/seriatim.normalize.stderr.log`
-  - `config/seriatim.normalize.generated.yml`
-
-## Trim Configuration
-
-`pipeline.trim` is optional. If omitted, no trim config is loaded. If `trim.enabled` is omitted, it defaults to `false`.
-
-When `trim.enabled: true`:
-
-- `trim.output_path` is required
-- `trim.bounds.prompt_id` is required
-- `trim.bounds.transcript_input_name` is required
-- `trim.bounds.output_path` is required
-- `trim.bounds.timeout` must be a valid Go duration when provided
-- `trim.bounds.render_debug: true` requires `trim.bounds.render_output_path`
-- `trim.bounds.profile_id` may be empty to use the prompt default profile
-
-Trim paths are treated as session-workdir-relative when not absolute.
-
-Example trim config:
-
-```yaml
-trim:
-  enabled: true
-  output_path: "transcripts/trimmed.json"
-  bounds:
-    prompt_id: "dnd_session.bounds"
-    profile_id: ""
-    transcript_input_name: "transcript"
-    output_path: "artifacts/session_bounds.json"
-    timeout: "10m"
-    render_debug: false
-    render_output_path: "artifacts/session_bounds.render.json"
-  seriatim:
-    report: false
+```sh
+narratio run 2026-04-04
 ```
 
-Trim behavior summary:
+This requires resolvable `pipeline.yml`, `campaign.yml`, and concrete
+`session.yml` files or their explicit command-line alternatives.
 
-- trim discovers and validates `transcripts/normalized.json`
-- trim uses Scriptorium bounds (`dnd_session.bounds` by example config) to produce `artifacts/session_bounds.json`
-- bounds IDs are validated against the same normalized transcript ID space that Seriatim trim will consume
-- trim converts bounds to Seriatim keep selector (for example `10-868`) and runs Seriatim trim
-- if trim is disabled, Narratio copies normalized transcript to trimmed transcript and records `trim_action=copy_disabled`
+## Documentation
 
-Trim outputs and diagnostics:
+- [CLI reference](docs/cli.md) — commands, arguments, flags, and invocation
+  behavior.
+- [Configuration](docs/config.md) — discovery, fields, defaults, and
+  validation.
+- [Operations](docs/operations.md) — runtime workflow, state, publishing,
+  recovery, and cleanup.
+- [Troubleshooting](docs/troubleshooting.md) — symptom-driven diagnosis and
+  safe remedies.
+- [Integration contracts](docs/integrations/) — external tools, formats, and
+  compatibility expectations.
+- [Maintained examples](examples/README.md) — complete copyable configuration
+  and input files.
 
-- `artifacts/session_bounds.json`
-- `transcripts/trimmed.json`
-- `logs/scriptorium.bounds.stdout.log`
-- `logs/scriptorium.bounds.stderr.log`
-- `config/scriptorium.bounds.generated.yml`
-- `logs/seriatim.trim.stdout.log`
-- `logs/seriatim.trim.stderr.log`
-- `config/seriatim.trim.generated.yml`
-- optional bounds render-debug outputs:
-  - `artifacts/session_bounds.render.json`
-  - `logs/scriptorium.bounds.render.stdout.log`
-  - `logs/scriptorium.bounds.render.stderr.log`
-  - `config/scriptorium.bounds.render.generated.yml`
+## Maintainer Documentation
 
-Render-debug files are diagnostics and are not treated as canonical stage output artifact refs.
-
-## Scriptorium Configuration
-
-`pipeline.scriptorium` is optional. When present, Narratio validates and uses it for analyze-stage artifact generation.
-
-Key points:
-
-- `scriptorium.binary` is required when section is present
-- `scriptorium.config_path` is optional
-- `scriptorium.timeout` defaults to `10m` when omitted
-- `scriptorium.render_debug` enables render diagnostics globally
-- artifacts are configured under `scriptorium.artifacts` (map shape supports multiple artifacts)
-- enabled artifacts require `prompt_id` and `output_path`
-- artifact `render_debug` may override global render setting
-- `vars` currently support boolean and string values
-
-Example `session_recap` artifact definition:
-
-```yaml
-scriptorium:
-  binary: "scriptorium"
-  config_path: "/etc/scriptorium/config.yml"
-  timeout: "10m"
-  render_debug: false
-
-  artifacts:
-    session_recap:
-      enabled: true
-      prompt_id: "dnd.session_recap"
-      profile_id: "local-quality" # optional
-      output_path: "artifacts/session_recap.md"
-      timeout: "10m"
-      # render_debug: true # optional per-artifact override
-
-      inputs:
-        transcript:
-          source: "trimmed_transcript"
-          required: true
-
-        previous_recap:
-          source: "previous_session_artifact"
-          artifact: "session_recap"
-          path: "" # optional; set when available
-          required: false
-
-      vars:
-        session_id: true
-        session_date: true
-        campaign_name: true
-        previous_session_id: true
-        output_kind: "session_recap"
-```
-
-Prompt IDs and profile IDs are configuration values. They are not hardcoded in analyze-stage logic.
-
-Do not put secrets in `pipeline.yml`. If API-key behavior is configured, use env var names only.
-
-If `pipeline.secrets.env_dir` is configured, keep only references and secret files there; secret values are still not written to manifests, generated configs, or Narratio-managed logs.
-
-## Scriptorium Runtime Behavior
-
-Narratio integrates with Scriptorium through the public CLI subprocess contract:
-
-- generation: `scriptorium run`
-- diagnostics/testing: `scriptorium render --format json` when `render_debug` is enabled
-
-For the initial implementation, only `session_recap` generation is supported.
-
-Analyze-stage session recap behavior:
-
-- available transcript input sources for configured artifacts: `processed_transcript`, `normalized_transcript`, `trimmed_transcript`
-- session recap should use gameplay-only transcript input (`source: trimmed_transcript`)
-- Narratio resolves `trimmed_transcript` from trim manifest output (`transcript_trimmed`) or fallback `transcripts/trimmed.json`
-- Narratio resolves `normalized_transcript` from normalize manifest output (`transcript_normalized`) or fallback `transcripts/normalized.json`
-- missing trimmed transcript fails clearly and advises running trim stage first
-- `normalized_transcript` is the preferred full-transcript source for future table/meta-analysis artifacts
-- `processed_transcript` remains supported for advanced/debug use cases
-- optionally includes `previous_recap` when configured and resolvable
-- omits optional previous recap when unavailable
-- fails if required inputs are missing
-- validates output file exists and is non-empty
-
-Expected session output paths:
-
-- `artifacts/session_recap.md`
-- `logs/scriptorium.session_recap.stdout.log`
-- `logs/scriptorium.session_recap.stderr.log`
-- `config/scriptorium.session_recap.generated.yml`
-- `artifacts/session_recap.render.json` when render diagnostics are enabled
-
-## Examples
-
-Starter files:
-
-- `examples/pipeline.minimal.yml`
-- `examples/session.minimal.yml`
-- `examples/speakers.yml`
-
-## Commands
-
-Run tests:
-
-```bash
-go test ./...
-```
-
-Plan a run:
-
-```bash
-go run ./cmd/narratio plan --session examples/session.minimal.yml
-```
-
-Use `--config <path>` to override default pipeline lookup when needed.
-
-Run full pipeline:
-
-```bash
-go run ./cmd/narratio run --config examples/pipeline.minimal.yml --session examples/session.minimal.yml
-```
-
-Run analyze only:
-
-```bash
-go run ./cmd/narratio run-stage --config examples/pipeline.minimal.yml --session examples/session.minimal.yml analyze
-```
-
-## Operational Note
-
-Checksum-based stale detection is not implemented yet.
-
-If prepared inputs or prompt/runtime config change, rerun the appropriate upstream stages before relying on downstream artifacts.
-
-Examples:
-
-- glossary/autocorrect/speaker-context changes: rerun at least `merge`, `polish`, `normalize`, `trim`, and `analyze`
-- trim bounds prompt/profile/config changes: rerun at least `normalize`, `trim`, and `analyze`
-- session recap prompt/profile/input-source changes: rerun `analyze`
-
-## Roadmap
-
-Near-term roadmap:
-
-- extend analyze to additional configured artifacts
-- support workflows where later artifacts consume earlier generated artifacts
-- keep orchestration explicit without a generic DAG engine
-- implement archive and notify backends
+- [Development guide](docs/development.md) — first-read orientation and
+  task-specific reading routes.
+- [Internal overview](docs/internal/overview.md) — implemented component map.
+- [Architecture](docs/policy/architecture.md) — normative boundaries and
+  invariants.
+- [Documentation policy](docs/policy/documentation.md) — canonical ownership
+  and maintenance rules.
+- [Testing policy](docs/policy/testing.md) — test value, boundaries, and
+  sufficiency.

architecture.md

@@ -1,358 +0,0 @@
-# Narratio Architecture
-
-## 1. Purpose
-
-`narratio` is a Go orchestrator for D&D session processing. It runs a stage-based local pipeline from audio input through transcript processing and artifact generation, with manifest-based skip/force/resume behavior.
-
-Narratio integrates with Scriptorium through the **public CLI** (`scriptorium run` and `scriptorium render`) via synchronous subprocess execution.
-
-## 2. Current Status
-
-Implemented:
-
-- strict `pipeline.yml` + `session.yml` loading with strict YAML field checking (`KnownFields(true)`)
-- local workspace/session layout, lock file handling, artifact path helpers, checksums, and atomic writes
-- manifest store and stage status transitions for resumable runs
-- real `prepare`, `transcribe`, `merge`, and `polish` stages
-- real WhisperX HTTP adapter
-- real Seriatim subprocess adapter
-- real Audita subprocess adapter
-- real Scriptorium subprocess adapter
-- real `normalize` stage producing `transcripts/normalized.json`
-- real `trim` stage producing `transcripts/trimmed.json`
-- real `analyze` stage for initial `session_recap` generation
-- optional Scriptorium render diagnostics (`render_debug`) before production run
-
-Still placeholder/future:
-
-- `archive` stage behavior
-- `notify` stage behavior
-- additional Scriptorium artifact types beyond `session_recap`
-- artifact-to-artifact workflows beyond the initial single-artifact implementation
-- generic stale detection based on input/config checksums
-
-## 3. Pipeline and Stage Boundaries
-
-Canonical stage order:
-
-1. `prepare`
-2. `transcribe`
-3. `merge`
-4. `polish`
-5. `normalize`
-6. `trim`
-7. `analyze`
-8. `archive`
-9. `notify`
-
-Boundary rules:
-
-- orchestration logic lives in `internal/app`
-- stage business logic lives in `internal/stage`
-- external-tool CLI construction lives in adapter packages
-- Scriptorium CLI details stay in `internal/adapters/scriptorium`
-
-## 4. Scriptorium Integration Model
-
-Integration mode:
-
-- public CLI subprocesses only (no Scriptorium internal Go packages, no HTTP API)
-- production generation uses `scriptorium run`
-- diagnostics/testing render uses `scriptorium render --format json`
-
-Run invocation shape used by adapter:
-
-```bash
-scriptorium run --prompt <prompt_id> --input name=path --out <output_path>
-```
-
-Optional flags passed when configured:
-
-- `--config <path>`
-- `--profile <profile_id>`
-- repeated `--var name=value`
-- repeated `--input name=path`
-- `--timeout <duration>`
-- `--api-key-env <ENV_NAME>` when configured
-
-Render invocation shape used by adapter:
-
-```bash
-scriptorium render --prompt <prompt_id> --input name=path --format json --out <render_output_path>
-```
-
-Adapter behavior:
-
-- always passes `--out`
-- captures stdout/stderr separately
-- writes generated invocation metadata YAML (redacted, no secrets)
-- treats exit code `0` as success
-- treats exit code `1` as failure
-- treats exit code `2` as failure with `validation_failed=true` and preserves output metadata when available
-- validates successful output files exist and are non-empty
-- does not treat non-empty stderr as failure by itself
-
-## 5. Configuration Contract
-
-CLI pipeline config path resolution:
-
-- when `--config <path>` is provided, that path is used
-- when `--config` is omitted, Narratio searches defaults in order:
-  - `/usr/local/etc/narratio/pipeline.yml`
-  - `/etc/narratio/pipeline.yml`
-
-Optional pipeline secrets directory:
-
-- `pipeline.secrets.env_dir` enables loading environment variables from local files before command execution
-- file name = env var name; file contents = env var value (trailing newline/CRLF trimmed)
-- only env-var-style file names are considered; other entries are ignored
-- existing process environment values are preserved (not overwritten)
-- if configured, unreadable/missing `env_dir` fails command execution early
-- relative `env_dir` values are resolved from current working directory
-
-`pipeline.scriptorium` is optional. Existing pipelines without Scriptorium continue to work.
-
-`pipeline.trim` is optional. Existing pipelines without trim config continue to work.
-
-`pipeline.normalize` is optional. Existing pipelines without normalize config continue to work.
-
-`pipeline.audita` drives the real Audita subprocess adapter for the `polish` stage.
-
-Audita required fields:
-
-- `binary`
-- `timeout`
-- `base_url`
-- `model`
-
-Audita optional fields:
-
-- `llm_api_key_env` (enforced only when configured)
-- `modules` override list (when omitted/empty, Narratio does not pass `--modules`)
-- `transcript_description`
-- `config_path`
-- `output_schema` (`bare-segments` or `audita-v1`)
-- `work_dir_retention` (`always`, `auto`, `never`)
-- `total_llm_concurrency` (> 0 when provided)
-- `proposal_llm_concurrency` (> 0 when provided)
-- `validation_model`
-- `validation_llm_concurrency` (> 0 when provided)
-- `report` (default `true`)
-
-Narratio passes only configured optional Audita flags; omitted optional values defer to Audita defaults.
-
-When `pipeline.normalize` is omitted, defaults are applied:
-
-- `output_path: transcripts/normalized.json`
-- `output_schema: seriatim-intermediate`
-- `report: true`
-
-When `pipeline.normalize` is present:
-
-- `output_path` must be non-empty
-- `output_schema` must be one of `seriatim-minimal`, `seriatim-intermediate`, or `seriatim-full`
-- relative `output_path` values are session-workdir-relative paths
-- Seriatim binary settings still come from `pipeline.seriatim`
-
-When `pipeline.trim` is present:
-
-- `enabled` is optional and defaults to `false` when omitted
-- relative `output_path`, `bounds.output_path`, and `bounds.render_output_path` values are session-workdir-relative paths
-- do not store secrets in trim config values
-
-When `pipeline.trim.enabled: true`:
-
-- `output_path` is required and non-empty
-- `bounds.prompt_id` is required and non-empty
-- `bounds.transcript_input_name` is required and non-empty
-- `bounds.output_path` is required and non-empty
-- `bounds.timeout` must parse as a Go duration when provided
-- `bounds.render_debug: true` requires non-empty `bounds.render_output_path`
-- `bounds.profile_id` may be empty to use the prompt default profile
-- prompt IDs are config values, not hardcoded stage logic
-
-When `pipeline.scriptorium` is present:
-
-- `binary` is required and non-empty
-- `config_path` is optional; when provided it must be non-empty
-- `timeout` is optional; when provided it must parse as a Go duration
-- default `timeout` is `10m`
-- unknown YAML fields fail strict decode
-
-Artifacts are configured as a map under `pipeline.scriptorium.artifacts` so multiple artifacts are possible in the config shape.
-
-For each artifact definition:
-
-- `enabled: true` requires non-empty `prompt_id`
-- `enabled: true` requires non-empty `output_path`
-- `timeout` must parse as Go duration when present
-- optional per-artifact `render_debug` may override global `scriptorium.render_debug`
-- `inputs` are named and each input requires non-empty `source`
-- inputs may be optional (`required: false`)
-- `vars` values currently support `string` and `bool`
-
-Prompt IDs and profile IDs are configuration values, not hardcoded stage logic.
-
-Trim config shape:
-
-```yaml
-trim:
-  enabled: true
-  output_path: "transcripts/trimmed.json"
-  bounds:
-    prompt_id: "dnd_session.bounds"
-    profile_id: ""
-    transcript_input_name: "transcript"
-    output_path: "artifacts/session_bounds.json"
-    timeout: "10m"
-    render_debug: false
-    render_output_path: "artifacts/session_bounds.render.json"
-  seriatim:
-    report: false
-```
-
-## 6. Transcript Tiers
-
-Narratio currently produces and uses four transcript tiers:
-
-- `transcripts/merged.json`: canonical deterministic merged transcript from Seriatim merge
-- `transcripts/processed.json`: full raw Audita-polished transcript output (includes pre/post-game content)
-- `transcripts/normalized.json`: normalized transcript generated by Seriatim normalize
-- `transcripts/trimmed.json`: gameplay-only normalized polished transcript from trim stage
-
-Trim reads `transcripts/normalized.json`, validates bounds IDs against that same transcript ID space, and writes `transcripts/trimmed.json`.
-
-## 7. Normalize Stage (Current Implementation)
-
-Normalize stage behavior:
-
-- stage order position: after `polish` and before `trim`
-- discovers processed transcript from manifest polish outputs (`transcript_processed`) when present, else `work/<session_id>/transcripts/processed.json`
-- validates processed transcript JSON shape (`segments` array required)
-- runs Seriatim `normalize` to produce normalized transcript
-- validates normalized transcript JSON shape (`segments` array required)
-- validates normalize report JSON when enabled
-
-Expected normalize outputs and diagnostics:
-
-- `transcripts/normalized.json`
-- `artifacts/seriatim.normalize.report.json` (when normalize report is enabled)
-- `logs/seriatim.normalize.stdout.log`
-- `logs/seriatim.normalize.stderr.log`
-- `config/seriatim.normalize.generated.yml`
-
-## 8. Trim Stage (Current Implementation)
-
-Trim stage behavior:
-
-- stage order position: after `normalize` and before `analyze`
-- discovers normalized transcript from manifest normalize outputs (`transcript_normalized`) when present, else `work/<session_id>/transcripts/normalized.json`
-- validates normalized transcript JSON shape (`segments` array required)
-- when `trim.enabled: false` (or trim config omitted), deterministically copies normalized transcript to `transcripts/trimmed.json` and records `trim_action=copy_disabled`
-- when `trim.enabled: true`:
-  - runs Scriptorium bounds prompt using configured `trim.bounds.prompt_id`
-  - writes bounds output to configured path (typically `artifacts/session_bounds.json`)
-  - parses and validates bounds output against the same normalized transcript being trimmed
-  - converts bounds range to Seriatim keep selector (for example `10-868`)
-  - runs Seriatim `trim` to produce `transcripts/trimmed.json`
-  - supports no-trim bounds actions (`none`/`copy`) by copying normalized transcript unchanged
-- validates trimmed transcript JSON shape (`segments` array required)
-
-Expected trim outputs and diagnostics:
-
-- `artifacts/session_bounds.json`
-- `transcripts/trimmed.json`
-- `logs/scriptorium.bounds.stdout.log`
-- `logs/scriptorium.bounds.stderr.log`
-- `config/scriptorium.bounds.generated.yml`
-- `logs/seriatim.trim.stdout.log`
-- `logs/seriatim.trim.stderr.log`
-- `config/seriatim.trim.generated.yml`
-- optional bounds render-debug outputs when enabled:
-  - `artifacts/session_bounds.render.json`
-  - `logs/scriptorium.bounds.render.stdout.log`
-  - `logs/scriptorium.bounds.render.stderr.log`
-  - `config/scriptorium.bounds.render.generated.yml`
-
-Render-debug files are diagnostics. They are recorded in stage metadata/log/config refs and are not treated as canonical stage output artifact refs.
-
-## 9. Analyze Stage (Current Implementation)
-
-The current real analyze implementation supports only `scriptorium.artifacts.session_recap`.
-
-Behavior:
-
-- if `pipeline.scriptorium` is missing, analyze returns a skipped result with metadata
-- if no Scriptorium artifacts are enabled, analyze returns a skipped result with metadata
-- if enabled artifacts exist but `session_recap` is not enabled, analyze fails clearly
-- available transcript input sources for configured artifacts: `processed_transcript`, `normalized_transcript`, `trimmed_transcript`
-- `session_recap` should use `trimmed_transcript` input (`transcripts/trimmed.json`) for in-universe recap generation
-- `trimmed_transcript` input is resolved from manifest (`trim` output kind `transcript_trimmed`) when available, otherwise fallback path `work/<session_id>/transcripts/trimmed.json`
-- `normalized_transcript` input is resolved from manifest (`normalize` output kind `transcript_normalized`) when available, otherwise fallback path `work/<session_id>/transcripts/normalized.json`
-- `processed_transcript` input is resolved from manifest (`polish` output kind `transcript_processed`) when available, otherwise fallback path `work/<session_id>/transcripts/processed.json`
-- `normalized_transcript` is the preferred full-transcript source for future table/meta-analysis artifacts
-- `processed_transcript` remains available for advanced/debug use cases
-- transcript inputs are validated as JSON with top-level `segments` array
-- configured inputs are resolved by source
-- optional `previous_recap` is omitted when unavailable
-- required `previous_recap` fails before invocation when unavailable
-- vars are built from config + session metadata
-- `render_debug` controls pre-run `scriptorium render` diagnostics
-- render failure stops stage before production run
-- render output is validated as JSON
-- production call uses Scriptorium adapter `RunArtifact`
-- successful run output must exist and be non-empty
-- missing `trimmed_transcript` input for configured `trimmed_transcript` source fails clearly with guidance to run trim stage first
-- manifest records output refs, logs, generated config paths, and non-secret provenance metadata
-
-## 10. Session Recap Paths
-
-Current expected paths for `session_recap`:
-
-- artifact output: `artifacts/session_recap.md`
-- run stdout log: `logs/scriptorium.session_recap.stdout.log`
-- run stderr log: `logs/scriptorium.session_recap.stderr.log`
-- run generated invocation/config: `config/scriptorium.session_recap.generated.yml`
-- render output (when enabled): `artifacts/session_recap.render.json`
-- render stdout log: `logs/scriptorium.session_recap.render.stdout.log`
-- render stderr log: `logs/scriptorium.session_recap.render.stderr.log`
-- render generated invocation/config: `config/scriptorium.session_recap.render.generated.yml`
-
-## 11. Security and Privacy
-
-- do not store secrets in pipeline YAML, generated invocation YAML, logs, or manifest metadata
-- if API-key integration is configured, pass env var names only (never raw key values)
-- with `pipeline.secrets.env_dir`, secret file values are loaded into process env only and are not persisted in manifest metadata or generated configs
-- avoid logging transcript content or rendered prompt content by default
-- treat generated artifacts and logs as potentially sensitive session material
-
-## 12. Operational Caveat (Pre-Stale-Detection)
-
-Checksum-based stale detection is not implemented yet.
-
-If prepared inputs or prompt/runtime configuration change (for example glossary files, prompt IDs, profile IDs, or relevant pipeline settings), rerun the appropriate prior stages to refresh downstream artifacts.
-
-Examples:
-
-- glossary or autocorrect changes usually require rerunning at least `merge`, `polish`, `normalize`, `trim`, and `analyze`
-- trim prompt/profile changes require rerunning at least `normalize`, `trim`, and `analyze`
-- session recap prompt/profile/input-source changes require rerunning `analyze`
-
-## 13. Roadmap
-
-Planned next steps:
-
-- extend analyze beyond `session_recap` to additional configured artifacts
-- support artifact inputs that consume prior generated artifacts
-- keep this composable without adding a generic DAG engine in the near term
-- implement real `archive` backend behavior
-- implement real `notify` backend behavior
-- add checksum-based stale detection and stale transitions
-
-Architectural invariants remain:
-
-- strict config decoding/validation
-- manifest-driven run control
-- clear stage/adapter separation
-- configuration-driven prompt/profile/input/vars/output mapping
-- Scriptorium integration through public CLI subprocess contract

docs/cli.md

@@ -0,0 +1,307 @@
+# CLI Reference
+
+## Shortest Useful Command
+
+```bash
+narratio run 2026-04-04
+```
+
+This runs the canonical full pipeline for session `2026-04-04`.
+
+## Command Overview
+
+Top-level commands:
+
+- `run <session_id>`: run full stage order.
+- `run-stage <stage> <session_id>`: run one stage.
+- `analyze <session_id>`: force-run analyze.
+- `publish <session_id>`: force-run publish.
+- `clean <session_id>` or `clean --all`: remove local work/spool state.
+- `session <subcommand>`: session helper commands.
+
+Session subcommands:
+
+- `session init <session_id>`
+- `session plan <session_id>`
+- `session validate <session_id>`
+- `session status <session_id>`
+- `session restore <session_id>`
+- `session artifacts <session_id>`
+- `session locks <session_id>`
+- `session locks add <session_id> <source>`
+- `session locks remove <session_id> <source>`
+
+## Common Config Flags
+
+Most session-aware commands accept:
+
+- `--config <pipeline.yml>`
+- `--campaign <id>`
+- `--campaign-file <campaign.yml>`
+- `--session <session.yml>`
+- `--session-id <session_id>`
+- `--previous-session-id <session_id>`
+
+Rules:
+
+- `--campaign` and `--campaign-file` are mutually exclusive.
+- `--session` is not used by `session init`.
+- if both positional `<session_id>` and `--session-id` are provided, values must match.
+- `--previous-session-id` is a strict expectation: the selected session file
+  must contain the same `previous_session_id`.
+- `clean --all` cannot be combined with campaign/session selectors.
+- notification delivery is currently limited to the configured `noop` mode; see
+  the [configuration reference](./config.md#notifications).
+
+## Session ID Input Rules
+
+Session-aware commands accept one of these forms:
+
+- positional session ID: `... <session_id>`
+- compatibility flag: `... --session-id <session_id>`
+
+When both are present, command parsing requires an exact match.
+
+Commands with additional positionals keep their command-specific order:
+
+- `run-stage <stage> <session_id>` or `run-stage <stage> --session-id <session_id>`
+- `session locks add <session_id> <source>` or `session locks add --session-id <session_id> <source>`
+- `session locks remove <session_id> <source>` or `session locks remove --session-id <session_id> <source>`
+
+## Command Reference
+
+### `run`
+
+```bash
+narratio run <session_id> [--force] [--artifacts <name[,name...]>] [...common config flags]
+```
+
+Behavior:
+
+- evaluates full stage order;
+- runs `extract` between `trim` and `render`; an omitted or disabled Notarius
+  configuration records an explicit `notarius_disabled` self-skip;
+- skips already-succeeded stages unless `--force` is set or a stage-specific
+  resume check finds its durable result obsolete;
+- continues interrupted or partially completed sessions by running non-succeeded stages;
+- writes session and run manifests.
+
+### `run-stage`
+
+```bash
+narratio run-stage <stage> <session_id> [--force] [--artifacts <name[,name...]>] [...common config flags]
+```
+
+Valid stage names:
+
+- `prepare`
+- `transcribe`
+- `merge`
+- `polish`
+- `normalize`
+- `trim`
+- `extract`
+- `render`
+- `analyze`
+- `publish`
+- `notify`
+
+Rules:
+
+- `--artifacts` is accepted only for `analyze` and `publish` stage targets.
+
+### `analyze`
+
+```bash
+narratio analyze <session_id> [--artifacts <name[,name...]>] [...common config flags]
+```
+
+Equivalent to:
+
+```bash
+narratio run-stage analyze <session_id> --force [...common config flags]
+```
+
+### `publish`
+
+```bash
+narratio publish <session_id> [--artifacts <name[,name...]>] [...common config flags]
+```
+
+Equivalent to:
+
+```bash
+narratio run-stage publish <session_id> --force [...common config flags]
+```
+
+### `clean`
+
+```bash
+narratio clean <session_id> [--dry-run] [--clear-cache] [...common config flags]
+narratio clean --all [--dry-run] [--clear-cache] [--config <pipeline.yml>]
+```
+
+Behavior:
+
+- session mode removes the selected session's local work and spool state;
+- `--all` removes all local session work and spool state;
+- cache remains unless `--clear-cache` is provided.
+
+See [Operations: Cleanup](./operations.md#cleanup) for deletion scope and
+post-publish cleanup behavior.
+
+### `session plan`
+
+```bash
+narratio session plan <session_id> [--force] [...common config flags]
+```
+
+Validates config, prepares local workdir layout, and prints run/skip decisions for each stage.
+
+### `session validate`
+
+```bash
+narratio session validate <session_id> [...common config flags]
+```
+
+Read-only preflight checks for config validity, required inputs, audio mode, previous-session requirements, publish outputs, and effective locks.
+
+### `session status`
+
+```bash
+narratio session status <session_id> [...common config flags]
+```
+
+Prints local manifest state and, when storage is available, status for the
+pointer-selected remote commit and its declared published outputs.
+
+### `session init`
+
+```bash
+narratio session init <session_id> --output ./session.yml [options]
+narratio session init <session_id> --remote [options]
+```
+
+Required target selection:
+
+- exactly one of:
+  - `--output <path>`
+  - `--remote`
+
+Options:
+
+- `--config <pipeline.yml>`
+- `--campaign <id>` or `--campaign-file <campaign.yml>`
+- `--previous-session-id <id>`
+- `--date <YYYY-MM-DD>`
+- `--title <text>`
+- `--audio-dir <path>`
+- `--audio-s3-prefix <prefix>`
+- `--force`
+
+Rules:
+
+- `--audio-dir` and `--audio-s3-prefix` are mutually exclusive.
+- if campaign `session_template_file` is configured, `session init` renders it.
+- generated session YAML must be concrete (no unresolved `{{ ... }}` placeholders).
+
+### `session restore`
+
+```bash
+narratio session restore <session_id> [--dry-run] [--force] [--include-audio] [...common config flags]
+```
+
+Behavior:
+
+- discovers committed remote current state;
+- plans local restores;
+- writes an execution report;
+- blocks unresolved conflicts. `--force` permits replacement only of eligible
+  regular files.
+
+See [Operations: Restore Workflow](./operations.md#restore-workflow) for the
+default restore scope, report location, and conflict-handling workflow.
+
+### `session artifacts`
+
+```bash
+narratio session artifacts <session_id> [--remote] [...common config flags]
+```
+
+Lists effective built-in, configured Scriptorium, and configured extraction
+sources; reports planned, available, unavailable, and published state without
+reading payload bodies; and includes publish rules, lock state, and optional
+remote published-state availability.
+
+### `session locks`
+
+```bash
+narratio session locks <session_id> [...common config flags]
+narratio session locks add <session_id> <source> [--reason <text>] [--force] [...common config flags]
+narratio session locks remove <session_id> <source> [...common config flags]
+```
+
+Behavior:
+
+- list mode reports the effective merge of static and remote locks;
+- add/remove mutate only remote locks;
+- static locks from pipeline config cannot be removed by CLI commands.
+
+See [Operations: Publish Locks](./operations.md#publish-locks) for lock storage
+and precedence.
+
+## `--artifacts` Selection Rules
+
+- accepted on `run`, `run-stage`, `analyze`, and `publish`;
+- names must exist in `pipeline.scriptorium.artifacts`;
+- empty entries are invalid;
+- repeated names are deduplicated.
+
+Effects:
+
+- filters analyze execution to selected configured artifacts;
+- filters publish rules that source `narratio.artifact.<name>`;
+- does not filter built-in transcript/bounds or explicitly configured
+  `narratio.extraction.<name>` publish sources; and
+- does not select or filter Notarius lanes.
+
+## Common Workflows
+
+Run full pipeline:
+
+```bash
+narratio run 2026-04-04
+```
+
+Dry-run restore plan:
+
+```bash
+narratio session restore 2026-04-04 --dry-run
+```
+
+Generate a concrete session file from template/default structure:
+
+```bash
+narratio session init 2026-04-04 --output ./session.yml --date 2026-04-04 --title "Session 12"
+```
+
+Force publish only:
+
+```bash
+narratio publish 2026-04-04
+```
+
+## Output And Exit Behavior
+
+- Successful commands write their result or summary to standard output and
+  exit with status `0`.
+- Command failures and invalid invocations write an error to standard error and
+  exit with status `1`.
+- An unknown top-level command also prints the top-level usage summary to
+  standard error.
+- `session restore --help` prints its command-specific usage and exits with
+  status `0`.
+
+Output is intended for operator inspection. Narratio does not currently offer
+a machine-readable CLI output mode; durable machine-readable state is recorded
+in manifests and reports described in [Operations](./operations.md).

docs/config.md

@@ -0,0 +1,406 @@
+# Configuration Reference
+
+## Purpose
+
+Narratio resolves three YAML documents:
+
+- `pipeline.yml`: pipeline/runtime settings
+- `campaign.yml`: campaign identity and stable input defaults
+- `session.yml`: session identity, metadata, and audio source selection
+
+## Discovery and Selection
+
+### `pipeline.yml`
+
+When `--config` is omitted, search order is:
+
+1. `/usr/local/etc/narratio/pipeline.yml`
+2. `/etc/narratio/pipeline.yml`
+
+### `campaign.yml`
+
+Selection rules:
+
+- if `--campaign-file` is set, use that path;
+- else if `--campaign <id>` is set, use `{pipeline.campaigns.root}/{id}/campaign.yml`;
+- else use `{pipeline.campaigns.root}/{pipeline.campaigns.default_campaign_id}/campaign.yml`.
+
+### `session.yml`
+
+When `--session` is omitted, local search order is:
+
+1. `/usr/local/etc/narratio/session.yml`
+2. `/etc/narratio/session.yml`
+
+If local session discovery fails and a `session_id` is known, Narratio attempts remote session loading from:
+
+- `{root_prefix}/campaigns/{campaign}/sessions/{session_id}/session.yml`
+
+using configured object storage.
+
+The downloaded remote session file is command-scoped: Narratio removes it after
+the command finishes and records only the remote object provenance alongside
+the durable copied session input.
+
+### Identity segments
+
+Campaign IDs (`campaign_id` and `default_campaign_id`), session IDs, previous
+session IDs, and Narratio run IDs are opaque portable segments. They must use
+only ASCII letters, digits, `.`, `_`, and `-`; empty values, `.`/`..`, path
+separators, drive forms, whitespace, control characters, and non-ASCII text are
+rejected. Narratio does not trim or rewrite these values. Existing manifests or
+remote state with an unsafe legacy identity must be migrated before use.
+
+## Validation and Merge Rules
+
+- YAML decode is strict (`KnownFields(true)`) and accepts exactly one document:
+  unknown fields or trailing documents fail load.
+- Configured timeout and retry-delay durations must be positive. An omitted
+  artifact timeout continues to inherit its configured Scriptorium timeout.
+- Session files must be concrete; unresolved `{{ ... }}` placeholders fail load.
+- Pipeline defaults are applied before validation.
+- Campaign and session identities must agree.
+- Required stable files (`speakers_file`, `autocorrect_file`, `glossary_file`,
+  `players_file`, `party_file`) and the optional `spell_catalog_file` resolve
+  from session overrides when provided, otherwise from campaign defaults. An
+  empty or omitted session spell-catalog value inherits the campaign value.
+- Exactly one audio mode must be configured in session input:
+  - local (`audio_dir` or `audio_files`), or
+  - S3 (`audio_s3.prefix`).
+
+## Minimal Working Configuration
+
+`pipeline.yml`
+
+```yaml
+campaigns:
+  root: /usr/local/share/narratio/campaigns
+  default_campaign_id: sample-campaign
+whisperx:
+  transcribe_url: https://transcription.example.com/transcribe
+```
+
+`campaign.yml`
+
+```yaml
+campaign_id: sample-campaign
+inputs:
+  speakers_file: ./speakers.yml
+  autocorrect_file: ./autocorrect.yml
+  glossary_file: ./glossary.yml
+  players_file: ./players.yml
+  party_file: ./party.yml
+```
+
+`session.yml` (local audio)
+
+```yaml
+session_id: 2026-05-03
+inputs:
+  audio_dir: ./audio
+```
+
+## Secrets Handling
+
+- Do not place raw secrets in YAML.
+- Use env var names in config (for example `pipeline.audita.llm_api_key_env`).
+- Optionally load credential files from `pipeline.secrets.env_dir`. Each valid
+  environment-variable filename supplies one value; trailing CR/LF is removed.
+- An existing process environment value takes precedence over a credential file.
+- Credential directories and files must not be symlinks and must be regular,
+  bounded files (at most 8 KiB per value). On POSIX, provision the directory
+  with no group/other access (normally `0700`) and files with no group/other
+  access (normally `0600`).
+- Commands that need storage/auth load filesystem secrets before constructing adapters.
+
+## Publish Configuration Summary
+
+Publish rules live under `pipeline.publish`.
+
+```yaml
+publish:
+  enabled: true
+  upload_run: true
+  outputs:
+    - source: narratio.transcript.final_trimmed
+      dest: transcripts/final.trimmed.json
+      required: true
+    - source: narratio.transcript.final_markdown
+      dest: transcripts/final.md
+      required: true
+    - source: narratio.transcript.final_trimmed_markdown
+      dest: transcripts/final.trimmed.md
+      required: true
+    - source: narratio.artifact.session_recap
+      dest: artifacts/session_recap.md
+      required: true
+  locks:
+    - source: narratio.artifact.session_recap
+      reason: manual post-publish edits
+```
+
+Rules:
+
+- `outputs[].source` is required.
+- `outputs[].dest` may be omitted when derivable from source.
+- extraction sources require an explicit `outputs[].dest` and publish only when
+  a rule names that source; the Notarius index and complete bundle are not
+  publish sources.
+- `outputs[].required` defaults to `true`.
+- static locks (`pipeline.publish.locks`) merge with remote locks (`{session_prefix}/locks.yml`), with static locks taking precedence on duplicates.
+
+## Full Schema
+
+### Pipeline
+
+| Field | Type | Required | Default / Rule |
+| --- | --- | --- | --- |
+| `pipeline.workspace.root` | string | No | `/var/lib/narratio` |
+| `pipeline.workspace.cleanup_after_publish` | bool | No | `false` |
+| `pipeline.campaigns.root` | string | No | `/usr/local/share/narratio/campaigns` |
+| `pipeline.campaigns.default_campaign_id` | string | No | empty |
+| `pipeline.secrets.env_dir` | string | No | empty |
+| `pipeline.storage.backend` | string | No | `local`; supported values are `local` and `s3` (case-insensitive) |
+| `pipeline.storage.s3.bucket` | string | Conditional | required when backend is `s3` and S3 session-audio or publish upload is enabled |
+| `pipeline.storage.s3.root_prefix` | string | No | `dnd` |
+| `pipeline.storage.s3.region` | string | No | empty |
+| `pipeline.storage.s3.endpoint` | string | No | empty |
+| `pipeline.storage.s3.force_path_style` | bool | No | `false` |
+| `pipeline.storage.s3.access_key_id_env` | string | No | `OBJECT_STORAGE_KEY_ID` |
+| `pipeline.storage.s3.secret_access_key_env` | string | No | `OBJECT_STORAGE_KEY` |
+| `pipeline.spool.root` | string | No | `/var/spool/narratio` |
+| `pipeline.spool.delete_audio_after_publish` | bool | No | `false` |
+| `pipeline.cache.root` | string | No | `/var/cache/narratio` |
+| `pipeline.cache.s3_audio` | bool | No | `true` |
+| `pipeline.publish.enabled` | bool | No | `true` |
+| `pipeline.publish.upload_run` | bool | No | `true` |
+| `pipeline.publish.outputs[]` | list | No | defaults to final trimmed JSON plus final and final-trimmed Markdown outputs |
+| `pipeline.publish.outputs[].source` | string | Yes (per rule) | must reference built-in or configured artifact source |
+| `pipeline.publish.outputs[].dest` | string | Conditional | derived if omitted and source supports derivation |
+| `pipeline.publish.outputs[].required` | bool | No | `true` |
+| `pipeline.publish.locks[]` | list | No | empty |
+| `pipeline.publish.locks[].source` | string | Yes (per lock) | must reference supported publish source |
+| `pipeline.publish.locks[].reason` | string | No | empty |
+| `pipeline.whisperx.transcribe_url` | string | Yes | absolute `http` or `https` URL |
+| `pipeline.whisperx.language` | string | No | `en` |
+| `pipeline.whisperx.timeout` | duration | No | `30m` |
+| `pipeline.whisperx.retries` | int | No | `3` |
+| `pipeline.whisperx.retry_delay` | duration | No | `2s` |
+| `pipeline.whisperx.concurrency` | int | No | `2` |
+| `pipeline.seriatim.binary` | string | No | `seriatim` |
+| `pipeline.seriatim.timeout` | duration | No | `10m` |
+| `pipeline.seriatim.output_schema` | string | No | `seriatim-intermediate` |
+| `pipeline.seriatim.coalesce_gap` | float | No | `3.0` |
+| `pipeline.seriatim.report` | bool | No | `true` |
+| `pipeline.seriatim.env.overlap_word_run_gap` | float | No | unset |
+| `pipeline.seriatim.env.overlap_word_run_reorder_window` | float | No | unset |
+| `pipeline.seriatim.env.backchannel_max_duration` | float | No | unset |
+| `pipeline.seriatim.env.filler_max_duration` | float | No | unset |
+| `pipeline.audita.binary` | string | No | `audita` |
+| `pipeline.audita.timeout` | duration | No | `3h` |
+| `pipeline.audita.llm_api_key_env` | string | No | empty |
+| `pipeline.audita.modules[]` | list[string] | No | empty |
+| `pipeline.audita.base_url` | string | No | empty |
+| `pipeline.audita.model` | string | No | empty |
+| `pipeline.audita.total_llm_concurrency` | int | No | unset |
+| `pipeline.audita.proposal_llm_concurrency` | int | No | unset |
+| `pipeline.audita.validation_model` | string | No | empty |
+| `pipeline.audita.validation_llm_concurrency` | int | No | unset |
+| `pipeline.audita.transcript_description` | string | No | empty |
+| `pipeline.audita.config_path` | string | No | empty |
+| `pipeline.audita.output_schema` | string | No | empty |
+| `pipeline.audita.work_dir_retention` | string | No | empty |
+| `pipeline.audita.report` | bool | No | `true` |
+| `pipeline.normalize.output_path` | string | No | `transcripts/final.json` |
+| `pipeline.normalize.output_schema` | string | No | `seriatim-intermediate` |
+| `pipeline.normalize.report` | bool | No | `true` |
+| `pipeline.trim.enabled` | bool | No | `true` |
+| `pipeline.trim.output_path` | string | No | `transcripts/final.trimmed.json` |
+| `pipeline.trim.bounds.prompt_id` | string | No | `dnd.session_bounds` |
+| `pipeline.trim.bounds.profile_id` | string | No | empty |
+| `pipeline.trim.bounds.transcript_input_name` | string | No | `transcript` |
+| `pipeline.trim.bounds.output_path` | string | No | `artifacts/session_bounds.json` |
+| `pipeline.trim.bounds.timeout` | duration | No | `10m` |
+| `pipeline.trim.bounds.render_debug` | bool | No | `false` |
+| `pipeline.trim.bounds.render_output_path` | string | Conditional | required when `render_debug` is true |
+| `pipeline.trim.seriatim.report` | bool | No | `false` |
+| `pipeline.notarius.enabled` | bool | No | `false` |
+| `pipeline.notarius.binary` | string | No | `notarius` |
+| `pipeline.notarius.config_path` | string | Conditional | required when enabled; relative paths resolve from the pipeline file directory |
+| `pipeline.notarius.pipeline_id` | string | Conditional | required when enabled |
+| `pipeline.notarius.timeout` | duration | No | `3h`; must be positive |
+| `pipeline.notarius.working_directory` | string | No | directory containing resolved `config_path`; relative paths resolve from the pipeline file directory |
+| `pipeline.notarius.references` | map[string]string | No | empty; maps normalized Notarius selectors to supported prepared Narratio source IDs; maximum 256 entries |
+| `pipeline.notarius.outputs` | map | Conditional | at least one entry when enabled |
+| `pipeline.render.enabled` | bool | No | `true` |
+| `pipeline.render.format` | string | No | `markdown` (only supported value) |
+| `pipeline.render.title` | string | No | empty (falls back to `session.title` when set) |
+| `pipeline.render.include_timestamps` | bool | No | `true` |
+| `pipeline.render.include_segment_ids` | bool | No | `true` |
+| `pipeline.render.include_metadata` | bool | No | `false` |
+| `pipeline.scriptorium.binary` | string | No | `scriptorium` |
+| `pipeline.scriptorium.config_path` | string | No | empty |
+| `pipeline.scriptorium.timeout` | duration | No | `10m` |
+| `pipeline.scriptorium.render_debug` | bool | No | `false` |
+| `pipeline.scriptorium.artifacts` | map | No | empty |
+| `pipeline.notification.mode` | string | No | `noop`; the only supported notification mode until a provider is implemented |
+
+### Notarius Reference Bindings
+
+`pipeline.notarius.references` maps a Notarius CLI selector to a prepared
+Narratio source, not to a filesystem path:
+
+```yaml
+notarius:
+  references:
+    glossary: narratio.input.glossary
+    party: narratio.input.party
+    players: narratio.input.players
+    spell_catalog: narratio.input.spell_catalog
+```
+
+Supported sources are `narratio.input.party`, `narratio.input.players`,
+`narratio.input.glossary`, and `narratio.input.spell_catalog`. Each map entry is
+required by its presence: omit a binding when the selected Notarius pipeline
+does not need it. A spell-catalog binding additionally requires an effective
+campaign or session `spell_catalog_file`.
+
+Selectors accept Notarius's `slot`, `chunk.slot`, `lane.slot`,
+`lane.extract.slot`, `lane.merge.slot`, and `lane.normalize.slot` forms.
+Narratio trims whitespace around
+selectors and their dot-separated components, rejects empty components and
+`=`, rejects duplicate normalized selectors, and limits the map to 256 entries.
+It validates only selector structure and the prepared source vocabulary;
+Notarius owns target-slot declarations and media compatibility.
+
+Before extraction, Narratio resolves every binding from the current prepared
+session manifest and streams it into a verified invocation-local snapshot whose
+absolute path is passed to Notarius. Missing, unsafe, empty,
+changed-during-copy, or checksum-inconsistent prepared evidence fails with
+guidance to force `prepare`. Bindings are sorted by normalized selector and are
+part of extraction fingerprint and resume identity. See the
+[Notarius integration contract](./integrations/notarius.md) for the subprocess
+boundary and the [complete example](../examples/pipeline.full.annotated.yml)
+for a copyable configuration.
+
+### Notarius Output Entries
+
+For each `pipeline.notarius.outputs.<name>`:
+
+| Field | Type | Required | Rule |
+| --- | --- | --- | --- |
+| `lane_id` | string | Yes | unique Notarius lane ID |
+| `media_type` | string | Yes | exact accepted descriptor media type |
+| `schema_id` | string | Yes | exact accepted descriptor schema ID |
+| `schema_version` | string | Yes | exact accepted descriptor schema version |
+| `module_key` | string | No | exact accepted module key when set |
+
+Output names must match `^[a-z][a-z0-9_]*$` and become selectable sources named
+`narratio.extraction.<name>`. Lane IDs must be unique. Every declared output is
+required from a successful Notarius result; a missing, rejected, duplicate, or
+contract-incompatible lane fails extraction. See the
+[complete maintained example](../examples/pipeline.full.annotated.yml) for the
+current ten-lane D&D mapping and the [Notarius contract](./integrations/notarius.md)
+for compatibility ownership.
+
+### Scriptorium Artifact Entries
+
+For each `pipeline.scriptorium.artifacts.<name>`:
+
+| Field | Type | Required | Rule |
+| --- | --- | --- | --- |
+| `enabled` | bool | No | `false` if omitted |
+| `depends_on[]` | list[string] | No | must reference configured artifact keys; no self-reference; enabled graph must be acyclic |
+| `render_debug` | bool | No | per-artifact override |
+| `prompt_id` | string | Conditional | required when artifact is enabled |
+| `profile_id` | string | No | empty |
+| `output_path` | string | Conditional | required when enabled; also required when referenced by publish/output/input rules |
+| `timeout` | duration | No | artifact override |
+| `inputs` | map | No | input key names must be non-empty |
+| `vars` | map | No | values must be string or bool; `session_id` is reserved and overwritten by Narratio |
+
+Narratio adds `session_id=narratio-session-<session_id>` to every Scriptorium request for sticky upstream LLM routing. If an artifact config sets `vars.session_id`, Narratio replaces that value before invoking Scriptorium. Use a different variable name if a prompt needs the raw Narratio session ID as content.
+
+Without `--artifacts`, analyze executes enabled configured artifacts. With an
+explicit `--artifacts` list, the exact named configured artifacts are the
+one-invocation execution set even if their `enabled` values are false; the list
+does not automatically include dependencies. Named artifacts must therefore be
+configured with valid executable fields, and their configured dependencies must
+already be available to analyze. This override affects analyze planning only;
+publish uses the list only to filter configured
+`narratio.artifact.<name>` output rules.
+
+For each artifact input `pipeline.scriptorium.artifacts.<name>.inputs.<input_name>`:
+
+| Field | Type | Required | Rule |
+| --- | --- | --- | --- |
+| `source` | string | Yes | built-in runtime source, prepared input source, `narratio.extraction.<name>`, `narratio.artifact.<name>`, or `narratio.previous_session.artifact.<name>` |
+| `required` | bool | No | optional input requirement |
+
+`artifact` and `path` are obsolete and rejected by strict configuration
+loading. Use the canonical `source` identifier to select the input; Narratio
+does not provide adapter-specific input passthrough fields.
+
+### Notifications
+
+Narratio currently supports only `notification.mode: noop`, which is also the
+default when the section is omitted. The notify stage performs no delivery in
+this mode. Backend, recipient, timeout, and other provider settings are
+rejected by strict configuration loading until Narratio has a provider
+integration.
+
+### Campaign
+
+| Field | Type | Required | Notes |
+| --- | --- | --- | --- |
+| `campaign_id` | string | Yes | canonical opaque campaign identity |
+| `session_template_file` | string | No | used by `session init` when set |
+| `inputs.speakers_file` | string | Yes | stable input default |
+| `inputs.autocorrect_file` | string | Yes | stable input default |
+| `inputs.glossary_file` | string | Yes | stable input default |
+| `inputs.players_file` | string | Yes | stable input default |
+| `inputs.party_file` | string | Yes | stable input default |
+| `inputs.spell_catalog_file` | string | No | optional spell-catalog overlay default; required when a Notarius reference selects `narratio.input.spell_catalog` |
+
+### Session
+
+| Field | Type | Required in session file | Notes |
+| --- | --- | --- | --- |
+| `session_id` | string | Yes | opaque identity; must match CLI session target when provided |
+| `previous_session_id` | string | No | opaque identity; must not equal `session_id` |
+| `campaign` | string | No | opaque identity; filled from `campaign_id` during resolve if omitted |
+| `date` | string | No | metadata |
+| `title` | string | No | metadata |
+| `inputs.speakers_file` | string | No | overrides campaign stable input |
+| `inputs.autocorrect_file` | string | No | overrides campaign stable input |
+| `inputs.glossary_file` | string | No | overrides campaign stable input |
+| `inputs.players_file` | string | No | overrides campaign stable input |
+| `inputs.party_file` | string | No | overrides campaign stable input |
+| `inputs.spell_catalog_file` | string | No | overrides the optional campaign spell catalog; empty or omitted inherits the campaign value |
+| `inputs.audio_dir` | string | Conditional | local audio mode |
+| `inputs.audio_files[]` | list[string] | Conditional | local audio mode |
+| `inputs.audio_s3.prefix` | string | Conditional | S3 audio mode |
+
+Audio rules:
+
+- configure local mode (`audio_dir` or `audio_files`) or S3 mode (`audio_s3.prefix`), not both.
+- `audio_s3` requires `pipeline.storage.backend: s3` and a configured S3 bucket.
+
+### Storage backend selection
+
+`local` is the default and disables remote object-store operations. Configure
+`s3` explicitly before supplying `storage.s3`; a populated S3 block does not
+select a backend on its own. Unknown backend names and an S3 block paired with
+`local` are rejected during configuration validation.
+
+### Previous-session expectation
+
+`previous_session_id` is optional in a session file. When a command supplies
+`--previous-session-id`, however, the session file must contain the same value;
+an omitted or different value is rejected before the command performs work.
+
+## Maintained Examples
+
+See the [maintained examples index](../examples/README.md) for complete pipeline,
+campaign, session, template, and input fixtures. Keep complete copyable files
+there rather than duplicating them in this reference.

docs/development.md

@@ -0,0 +1,58 @@
+# Development
+
+This is the first-read landing page for people and LLM coding agents working on
+Narratio. It provides a concise repository orientation and routes each kind of
+change to its canonical documentation.
+
+Narratio is a stage-driven Go orchestrator for turning D&D session audio into
+polished transcripts and generated artifacts. Start with the
+[README](../README.md) for product context,
+[Architecture](policy/architecture.md) for normative system boundaries, and the
+[Internal Overview](internal/overview.md) for implemented component ownership.
+
+## What To Read
+
+| When working on | Read | Why |
+| --- | --- | --- |
+| Finding the package or component that owns current behavior | [Internal Overview](internal/overview.md) | It is the implemented component inventory and routes to focused internal documents. |
+| Application shape, boundaries, dependency direction, runtime invariants, safety properties, or dependencies | [Architecture](policy/architecture.md) | It defines the intended system shape, ownership, and non-goals. |
+| Any documentation addition or revision | [Documentation Policy](policy/documentation.md) | It defines canonical owners, audiences, current-behavior rules, and maintenance requirements. |
+| Adding, changing, reviewing, rewriting, or deleting tests | [Testing Policy](policy/testing.md) | It defines risk-based sufficiency, durable test boundaries, test-double guidance, and test lifecycle decisions. |
+| CLI composition or command behavior | [Internal Overview](internal/overview.md) and [CLI Reference](cli.md) | The overview routes to command ownership; the reference owns public syntax and invocation behavior. |
+| Configuration loading, resolution, or user-visible configuration | [Internal Overview](internal/overview.md) and [Configuration](config.md) | The overview routes to implementation ownership; the reference owns fields, defaults, discovery, and validation. |
+| Session workflow, status, restore, cleanup, or object storage | [Restore Internals](internal/command-restore.md), [Workspace Internals](internal/workspace.md), [Storage Internals](internal/storage.md), [Operations](operations.md), and [Troubleshooting](troubleshooting.md) | These separate implementation mechanics, operator procedures, and symptom-driven recovery. |
+| Pipeline sequencing or the behavior of a stage | [Internal Overview](internal/overview.md) and its focused stage documents | The overview owns the implemented stage inventory and routes to each stage contract. |
+| Adapters or external tool contracts | [Adapter Internals](internal/adapters.md) and [Integration Contracts](integrations/README.md) | The internal guide owns adapter composition and mechanics; integration documents own external formats and protocols. |
+| Manifests, artifacts, workspace paths, or publish behavior | [Manifest Internals](internal/manifest.md), [Artifact Internals](internal/artifacts.md), [Workspace Internals](internal/workspace.md), [Publish Internals](internal/stage-publish.md), and [Operations](operations.md) | These separate implementation state and resolution from operator-visible layout and lifecycle. |
+| Maintained configuration or input examples | [Configuration](config.md) and [Examples](../examples/README.md) | The reference owns field meanings; the examples directory owns complete copyable files. |
+| Proposed or unimplemented behavior | `docs/roadmap/` | Future work belongs only in roadmap documentation until implemented. |
+
+For an existing subsystem, also inspect its focused tests and package-level
+contracts before changing behavior.
+
+## Validation
+
+Use focused package tests while iterating. Every pull request and push runs the
+following repository-wide checks before it can be accepted:
+
+```sh
+go test ./...
+go test -race ./...
+go vet ./...
+go build ./...
+go test ./internal/doccheck
+go test ./internal/config -run '^TestExamplesLoadAndValidate$'
+```
+
+The documentation check verifies local Markdown links and the dependency graph
+of the Woodpecker workflows. The configuration check loads every maintained
+pipeline and session example. Release automation repeats these checks and
+cross-compiles the CLI before it builds release assets; publishing depends on
+that validation path, so a failure cannot publish a release.
+
+Woodpecker also runs `go test -race -shuffle=on -count=3 ./...` on its scheduled
+job to expose ordering and repeatability defects. Current runners cross-compile
+for macOS and Windows, but do not provide native macOS or Windows execution.
+Those cross-builds establish compilation only, not platform-equivalent runtime
+evidence. Add native checks only when official runner labels and successful
+native-run evidence are available.

docs/integrations/README.md

@@ -0,0 +1,35 @@
+# Integrations Index
+
+## Audience
+
+Operators, developers, and coding agents who need to understand Narratio's
+externally observable integration boundaries.
+
+## Scope
+
+`docs/integrations/` is the canonical reference for protocols, invocation and
+data contracts, logical outputs, and compatibility behavior at external tool
+boundaries.
+
+These documents describe what Narratio sends or invokes, what it accepts in
+return, and how failures are surfaced. Internal composition and stage mechanics
+belong in [the adapter implementation guide](../internal/adapters.md) and the
+focused stage documents.
+
+## Integration Contracts
+
+- [Audita](./audita.md): transcript polishing (`audita process`).
+- [Notarius](./notarius.md): complete pipeline execution and safe JSON bundle
+  discovery (`notarius run`).
+- [Seriatim](./seriatim.md): merge, normalize, trim, and render operations.
+- [Scriptorium](./scriptorium.md): artifact generation and debug rendering
+  (`scriptorium run|render`).
+- [WhisperX](./whisperx.md): speaker-audio transcription over HTTP.
+
+## Related Canonical Docs
+
+- [Configuration](../config.md): operator-facing configuration reference.
+- [Adapter implementation](../internal/adapters.md): shared adapter boundary and
+  runner wiring.
+- [Internal documentation](../internal/overview.md): stage-specific integration
+  usage and component ownership.

docs/integrations/audita.md

@@ -1,96 +1,67 @@
-# Audita Subprocess Operations
+# Integration: Audita
 
-This document describes how parent processes should invoke `audita process` safely in production orchestration.
+## Purpose
+Define the Audita adapter contract used by the `polish` stage.
 
-## Recommended command form
+## External Boundary
 
-Use explicit file outputs for orchestrated runs:
+Narratio invokes `audita process` as a subprocess for each polish operation.
+The configured timeout and parent cancellation bound the invocation. Internal
+runner composition is documented in
+[the adapter implementation guide](../internal/adapters.md).
 
-```sh
-audita process <transcript.json> \
-  --transcript-description "Brief context that may help resolve ambiguous terms." \
-  --glossary <glossary.yaml> \
-  --output <output-transcript.json> \
-  --report-json <report.json>
-```
+## Request Contract
+`PolishRequest` carries:
+- required transcript/glossary/output/work-dir paths;
+- optional report path (required when report mode is enabled);
+- generated config and stdout/stderr log paths;
+- optional per-invocation module override.
 
-Additional flags that may be situationally appropriate:
-- `--config <path>` to select an explicit versioned config file.
-- `--output-schema <bare-segments|audita-v1>` to select transcript output shape.
-- `--work-dir <dir>` to control diagnostics location.
-- `--work-dir-retention <always|auto|never>` to control retained run directories.
-- `--total-llm-concurrency`, `--proposal-llm-concurrency`, and `--validation-llm-concurrency` when orchestration needs to set explicit LLM throughput controls.
-- `--modules ...` only when intentionally overriding the default sequence.
+The constructed runner owns static Audita settings: binary, timeout,
+credentials, default modules, model and endpoint settings, validation and output
+settings, report mode, and concurrency. The `polish` stage supplies only
+invocation-specific paths and may override modules for that invocation.
 
-For config-driven orchestration, validate config files in CI/preflight:
+## Result Contract
+`PolishResult` returns:
+- processed transcript path;
+- optional report path;
+- work dir and generated-config/log paths;
+- exit code, duration, binary provenance;
+- adapter metadata map.
 
-```sh
-audita config validate --config <path>
-```
+## Validation and Failure Semantics
+Construction fails for invalid static config values, including:
+- empty binary;
+- non-positive timeout;
+- invalid base URL;
+- invalid output schema;
+- invalid work-dir retention value;
+- invalid concurrency values.
 
-## Stdout behavior
+Run fails for:
+- missing required request paths;
+- missing required credential env var when configured (`llm_api_key_env`);
+- subprocess execution failure;
+- invalid processed transcript JSON (`segments` array required);
+- invalid report JSON when reporting is enabled.
 
-- With `--output`: stdout is expected to be empty on success.
-- Without `--output`: stdout contains transcript JSON only on success.
-- Report JSON is never written to stdout.
+Processed transcript JSON is limited to 64 MiB and optional report JSON to 16
+MiB. Both must be regular files without symlinked path components.
 
-## Stderr behavior
+Failure results still include output/log/config/exit metadata for diagnostics.
 
-- Success path should be quiet or minimal human-readable logs.
-- Failure path writes concise human-readable errors.
-- When a diagnostics run directory exists, failure stderr includes its path.
-- Prompt/response diagnostic payloads are not streamed to stderr.
+## Deterministic Behavior
+- CLI args are built from runner config + request in a fixed order.
+- Generated invocation YAML (`audita.generated.v1`) is emitted when requested.
+- Manifest writes are stage-owned; adapter itself is stateless.
 
-## Output file behavior
+## Configuration
 
-- `--output` writes transcript JSON in the selected output schema to the provided path.
-- Output write failures return nonzero and surface actionable errors.
-- The command does not silently ignore output write errors.
+Operator-selected values are defined under `pipeline.audita.*` in the
+[configuration reference](../config.md#pipeline).
 
-## Report JSON behavior
+Maintained example with Audita config:
 
-- `--report-json` writes a machine-readable process report to the requested path.
-- Run-directory `report.json` is written independently under diagnostics.
-- Best-effort failure reports are emitted when possible without masking the primary failure.
-- Report write failures return nonzero with clear stderr messaging.
-- Report diagnostics metadata references run-directory artifacts including utilization diagnostics and correction ledger paths when available.
-
-## Diagnostics directory behavior
-
-- Each run creates (when possible) a per-run diagnostics directory.
-- Typical artifacts include transcript, normalization, chunking, invocation, effective config, LLM diagnostics, `utilization-diagnostics.json`, `correction-ledger.json`, `report.json`, and `error.log` on failure.
-- Failed runs retain diagnostics.
-- Under `auto` retention, successful runs with skipped/rejected corrections are retained; clean successful runs may be removed.
-
-## Exit codes
-
-- `0`: success.
-- Nonzero: failure (input/schema/config/module/LLM/runtime/output/report/diagnostics errors).
-
-Treat any nonzero as a failed subprocess invocation.
-
-## Timeout and cancellation
-
-- Runtime operations propagate context cancellation and request timeouts through LLM/scheduler paths.
-- On cancellation or timeout, the process exits nonzero and should not hang.
-- If diagnostics were initialized before failure, failure artifacts remain available for debugging.
-
-## Secret redaction expectations
-
-API keys and configured secret values are redacted from:
-- reports (`--report-json` and run-dir `report.json`);
-- diagnostics artifacts (including effective config and LLM interaction artifacts);
-- surfaced adapter/runtime errors;
-- test fixtures and regression outputs.
-
-Parent-process logs should still avoid printing raw environment variables.
-
-## Parent-process pipe guidance
-
-To avoid deadlocks in orchestrators:
-- always read both stdout and stderr concurrently when invoking as a subprocess;
-- prefer file outputs (`--output`, `--report-json`) for machine workflows;
-- treat stderr as human-readable diagnostics, not structured data;
-- parse structured results from output/report files.
-
-For Go callers, prefer `exec.CommandContext` with explicit timeout/cancellation and buffered/streamed readers for both pipes.
+- [Full annotated pipeline](../../examples/pipeline.full.annotated.yml)
+- [Production-shaped pipeline](../../examples/pipeline.production.yml)

docs/integrations/notarius.md

@@ -0,0 +1,126 @@
+# Notarius Integration Contract
+
+## Boundary
+
+Narratio uses Notarius as a subprocess to extract configured structured JSON
+lanes from the final trimmed Seriatim transcript. Narratio owns invocation,
+safe bundle discovery, lane selection, and its own artifact metadata. Notarius
+owns pipeline definitions, lane schemas, the receipt, and bundle formats.
+
+Canonical Notarius v0.6.0 references:
+
+- [CLI reference](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/cli.md)
+- [Subprocess consumer contract](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/consumers/subprocess.md)
+- [D&D pipeline and lane contracts](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/consumers/dnd-pipeline.md)
+- [Run-result receipt](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/integrations/run-result.md)
+- [JSON output bundle](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/integrations/json-output.md)
+- [D&D spell-catalog overlay](https://gitea.maximumdirect.net/eric/notarius/src/tag/v0.6.0/docs/integrations/dnd-spell-catalog-overlays.md)
+
+The [complete Narratio example](../../examples/pipeline.full.annotated.yml)
+records the exact current constraints for all ten D&D lanes. Treat the linked
+Notarius documents as canonical when changing those values; Narratio does not
+duplicate the complete schemas.
+
+## Invocation
+
+When `pipeline.notarius.enabled` is true, Narratio resolves the executable,
+configuration path, input path, output directory, and working directory to
+absolute paths. Narratio requires the Notarius v0.6.0 CLI contract when
+references are configured and invokes each binding as a separate argument
+before `--json`:
+
+```text
+notarius run <pipeline_id> --config <config_path> --input <trimmed_json> --output-dir <staging_dir> [--reference <selector>=<verified_snapshot_path>]... --json
+```
+
+Reference paths are absolute invocation-local snapshots streamed from the
+manifest-verified canonical files prepared inside the current Narratio session
+workspace. Narratio verifies snapshot checksum and size before and after the
+subprocess, and passes only configured bindings, ordered lexically by normalized
+selector, as direct argument-vector entries without shell interpretation. A CLI
+binding takes precedence over a matching external path in Notarius
+configuration. Narratio never emits `--without-reference`.
+
+The maintained D&D boundary binds only the four campaign-owned external slots:
+
+```text
+notarius run dnd-session \
+  --config <absolute config path> \
+  --input <absolute trimmed transcript path> \
+  --output-dir <absolute staging directory> \
+  --reference glossary=<absolute verified glossary snapshot> \
+  --reference party=<absolute verified party snapshot> \
+  --reference players=<absolute verified players snapshot> \
+  --reference spell_catalog=<absolute verified spell catalog snapshot> \
+  --json
+```
+
+The spell-catalog binding is omitted when the campaign does not maintain that
+optional overlay. Registry, scene-description, combat-turn, and occurrence
+handoffs generated during the same Notarius run remain in Notarius pipeline
+composition and must not be emitted as CLI references. The linked CLI and D&D
+consumer documents own selector targeting, declared slots, media compatibility,
+and generated-handoff collision rules.
+
+Standard output is reserved for the JSON receipt. Standard error is captured
+separately as diagnostic output. Narratio applies the configured timeout and
+does not interpret stdout as a receipt unless the subprocess exits successfully.
+It does not pass a Narratio session ID or run `notarius config validate`
+automatically; the configured working directory and Narratio's minimal child
+environment apply to the subprocess.
+
+## Accepted Result
+
+Narratio's supported invocation baseline is Notarius v0.6.0. The accepted
+receipt remains `notarius.run-result.v2`; reference flags do not change the
+receipt or ten-lane output contract. The receipt
+must identify the configured pipeline, and its `index_file` must be exactly
+`index.json` beneath the reported bundle root. The production index must name
+the management files exactly as `manifest.json`, `rejected.json`,
+`warnings.json`, and `diagnostics.json`. All receipt, index, and lane paths must
+stay inside that bundle; symlinks and non-regular lane payloads are rejected.
+
+Supported receipt and index shapes tolerate unknown fields for forward
+compatibility, while required identity, validation, count, manifest,
+rejection, warning, diagnostic, and lane-list fields remain mandatory.
+Narratio applies bounded reads to the receipt, index, rejection, warning, and
+diagnostic documents. Warning and diagnostic envelopes, group counts,
+occurrence counts, truncation state, framework-owned origins, and
+receipt-to-bundle counts must be internally consistent. Optional chunk-map and
+evidence-context descriptors must carry their complete generic contract
+metadata when present.
+
+For every entry in `pipeline.notarius.outputs`, Narratio requires exactly one
+index descriptor with the configured lane ID, media type, schema ID, schema
+version, and, when configured, module key. Missing, duplicate, rejected, or
+incompatible required lanes fail extraction even if Notarius exited zero. A
+configured lane whose v2 validation summary is `rejected` or `incomplete` also
+fails extraction.
+Unconfigured lanes may remain in the preserved bundle but do not become
+selectable Narratio sources.
+
+Each accepted configured lane is registered as
+`narratio.extraction.<output_key>`. The bundle index is retained for audit and
+resume validation but is not selectable. Scriptorium and publish rules consume
+only explicitly named lane sources; `--artifacts` never selects Notarius lanes.
+
+## Failure And Compatibility Behavior
+
+- Startup and nonzero-exit errors fail extraction and retain captured diagnostics.
+- Invalid receipt JSON or an unsupported receipt schema fails before bundle use.
+- Unsafe or incompatible index data and required-lane rejection fail before the
+  staged bundle is promoted to durable storage.
+- Contract and external provenance metadata are preserved on lane artifact
+  records and through explicit publication.
+- Undeclared selectors, incompatible reference files, and external/generated
+  reference collisions are Notarius errors and fail extraction normally.
+
+Rejection, validation, warning, and diagnostic summaries retain bounded stable
+identity, category, origin, reason-code, status, and occurrence fields without
+copying free-form external messages into Narratio manifest metadata or reading
+lane payload bodies.
+
+Configuration fields and defaults are in [Configuration](../config.md).
+Operator paths, rerun procedures, and bundle retention are in
+[Operations](../operations.md). See [Troubleshooting](../troubleshooting.md)
+for failure recovery.

docs/integrations/scriptorium.md

@@ -1,339 +1,71 @@
-# Narratio -> Scriptorium CLI Integration
+# Integration: Scriptorium
 
-## 1. Purpose
+## Purpose
+Define the Scriptorium adapter contract used by `analyze` and trim-bounds generation in `trim`.
 
-This document defines how Narratio should invoke Scriptorium through the **public CLI**.
+## External Boundary
 
-This is a **subprocess integration contract**, not an internal Go API contract.
-
-## 2. Assumptions
-
-- `scriptorium` is installed and available on `PATH`.
-- Scriptorium is configured with `config.yml`.
-- `config.yml` provides `prompt_dir`, `profile_dir`, and `schema_dir` as needed.
-- Prompt and profile libraries are already deployed for the environment.
-- Narratio provides prepared artifact files (for example polished transcript, glossary, previous recap, campaign notes).
-- Initial integration is synchronous subprocess execution.
-- Narratio remains the orchestrator.
-
-In normal operation, Narratio does not need to pass `--prompt-dir` and `--profile-dir` if they are supplied by Scriptorium config.
-
-Narratio may pass `--config <PATH>` when it must use a non-default Scriptorium config file.
-
-## 3. Core Commands Narratio May Call
-
-Primary commands for subprocess integration:
+Narratio invokes Scriptorium as a subprocess in these modes:
 
 - `scriptorium run`
 - `scriptorium render`
 
-For production generation, use `scriptorium run`.
-
-`scriptorium render` is for debugging, dry-runs, test assertions, and validating command construction without LLM execution.
-
-Note: `scriptorium serve` and HTTP API exist, but they are not the initial integration path.
-
-## 4. Command Selection Guidance
-
-- Use `run` to generate an output artifact.
-- Use `render` to inspect the prepared prompt and effective settings without calling the LLM.
-- Use `render --format json` when Narratio/tests need structured prepare output.
-
-## 5. Recommended `run` Invocation Shape
-
-Production shape:
-
-```bash
-scriptorium run \
-  --prompt <prompt_id> \
-  --input transcript=<processed-transcript-path> \
-  --out <output-artifact-path>
-```
-
-Common optional additions:
-
-- `--config <path>`: use a specific Scriptorium config file.
-- `--profile <profile_id>`: override prompt default profile.
-- `--var name=value` (repeatable): small metadata values.
-- `--input name=path` (repeatable): additional named artifacts.
-- `--timeout <duration>`: per-run timeout override.
-- Runtime model override flags (`--llm-base-url`, `--model`, etc.) only for exceptional/operator-directed cases.
-
-## 6. Recommended `render` Invocation Shape
-
-Human-readable debug shape:
-
-```bash
-scriptorium render \
-  --prompt <prompt_id> \
-  --input transcript=<processed-transcript-path> \
-  --format text
-```
-
-Structured debug/test shape:
-
-```bash
-scriptorium render \
-  --prompt <prompt_id> \
-  --input transcript=<processed-transcript-path> \
-  --format json \
-  --out <render-debug-path>
-```
-
-`render` does **not** call the LLM, does **not** validate model output, and does **not** perform repair.
-
-## 7. Inputs
-
-- Pass inputs as repeated `--input name=path` flags.
-- `name` must match the Prompt Definition input name.
-- Prefer absolute paths, or paths relative to a working directory controlled by Narratio.
-- Pass Audita output as the primary transcript input.
-- Additional inputs may include glossary, previous recap, campaign notes, event logs, final state maps, or other prompt-specific artifacts.
-- Scriptorium reads input files directly; Narratio does not need to inline file content for CLI use.
-
-## 8. Variables
-
-Use repeated `--var name=value` for small metadata values.
-
-Typical examples:
-
-- `session_date`
-- `session_id`
-- `campaign_name`
-- `previous_session_id`
-- `output_kind`
-
-Large content belongs in input files, not `--var` values.
-
-## 9. Prompt IDs and Output Artifact Types
-
-Narratio should treat prompt IDs as configuration, not hardcoded business logic.
-
-Narratio config may map stage/output names to prompt IDs, for example:
-
-- session recap prompt
-- structured event extraction prompt
-- glossary suggestion prompt
-- player-facing summary prompt
-
-Prompt IDs used by Narratio should come from the deployed Scriptorium prompt library.
-
-## 10. Profiles
-
-- Prompts may declare `default_profile`.
-- Narratio may omit `--profile` to use prompt default profile.
-- Narratio may pass `--profile` to force profile selection.
-- This enables environment/profile selection like `local-fast`, `local-quality`, `frontier`, `batch`, or test profiles.
-- Profile names should generally be Narratio configuration values.
-
-## 11. Runtime Overrides
-
-Supported runtime override flags:
-
-- `--llm-base-url`
-- `--model`
-- `--api-key-env`
-- `--temperature`
-- `--max-tokens`
-- `--top-p`
-- `--timeout`
-
-Guidance:
-
-- Keep normal model/runtime settings in Execution Profiles.
-- Use runtime overrides only for explicit per-run exceptions, tests, or operator overrides.
-- Never pass raw API keys on the command line.
-- `--api-key-env` names an environment variable; Narratio must ensure that variable is set in subprocess environment.
-
-## 12. Config Behavior
-
-- Default config path: `/etc/scriptorium/config.yml`.
-- `--config <PATH>` overrides default path.
-- Missing default config is allowed by Scriptorium.
-- If `--config` is provided explicitly, the file must exist and be valid.
-- CLI flags override `config.yml`.
-- `config.yml` overrides built-in application defaults.
-
-Narratio can either:
-
-- rely on system default config path, or
-- carry an explicit config path and pass `--config`.
-
-## 13. Environment Handling
-
-Subprocess environment recommendations:
-
-- Pass through required API-key environment variables referenced by `api_key_env`.
-- Do not pass raw API keys as CLI arguments.
-- Avoid logging full environment dumps.
-- Capture stdout and stderr separately.
-- Use a controlled working directory.
-- Prefer absolute artifact paths.
-
-## 14. Output Handling
-
-For `scriptorium run`:
-
-- Use `--out` when Narratio needs durable artifact files.
-- Without `--out`, artifact content is written to stdout.
-- Preferred orchestration pattern: always use `--out`, then treat the file as stage output artifact.
-- Capture stderr for diagnostics.
-
-For `scriptorium render`:
-
-- Use `--out` to store render diagnostics.
-- Use `--format json` when tests need to inspect selected profile, effective runtime settings, input hashes, prompt hash, and rendered messages.
-
-## 15. Exit Status and Errors
-
-Current CLI behavior (verified from implementation/tests):
-
-- `0`: success.
-- `1`: runtime/parse/config/load/render/generation/IO error.
-- `2`: run completed but output validation failed (`ValidationFailed`).
-
-Additional details:
-
-- On `run`, output artifact write happens before exit code selection. If validation fails, artifact may still be written and exit code is `2`.
-- `stderr` carries both errors and normal run summary output; non-empty stderr alone does not imply failure.
-- `render` returns `0` on success and `1` on failures.
-
-Narratio should treat non-zero exit codes as failed stage execution, but may record generated artifact paths if a run exited `2` and output file exists.
-
-## 16. Recommended Narratio Integration Pattern
-
-1. Build CLI args from Narratio stage configuration.
-2. Use subprocess context cancellation/timeout.
-3. Pass absolute input paths.
-4. Pass `--out` to a session-scoped artifact path.
-5. Add `--var` metadata values.
-6. Optionally add `--config`.
-7. Optionally add `--profile`.
-8. Ensure required API-key env vars are present.
-9. Run subprocess synchronously.
-10. Capture stdout/stderr separately.
-11. On success, store output artifact path and invocation metadata in stage artifacts.
-12. On failure, store exit code and stderr diagnostics in stage status.
-
-## 17. Suggested Narratio Configuration Shape
-
-Illustrative `pipeline.yml` shape:
-
-```yaml
-scriptorium:
-  binary: scriptorium
-  config_path: /etc/scriptorium/config.yml
-  timeout: 10m
-  render_debug: false
-  artifacts:
-    session_recap:
-      enabled: true
-      prompt_id: dnd.session_recap
-      profile_id: local-quality   # optional
-      output_path: artifacts/session_recap.md
-      timeout: 10m
-      render_debug: false          # optional artifact override
-      inputs:
-        transcript:
-          source: trimmed_transcript
-          required: true
-        previous_recap:
-          source: previous_session_artifact
-          artifact: session_recap
-          path: ""                 # optional
-          required: false
-      vars:
-        session_id: true
-        session_date: true
-        campaign_name: true
-        previous_session_id: true
-        output_kind: session_recap
-```
-
-The key idea: map Narratio artifact names to prompt ID, optional profile, expected inputs, vars, and output destination.
-
-## 18. Testing Strategy for Narratio Integration
-
-- Use `scriptorium render --format json` to verify command construction without LLM calls.
-- Use dedicated test prompt/profile libraries for integration tests.
-- Use small fixture transcripts.
-- Verify missing-input failure behavior.
-- Verify prompt `default_profile` behavior.
-- Verify explicit `--profile` override behavior.
-- Verify `--config` behavior (default and explicit).
-- Verify output file creation when `--out` is used.
-- Verify stderr capture on failures.
-- Avoid real API keys in tests.
-
-## 19. Security and Privacy Notes
-
-- Never pass raw API keys on command line.
-- Do not log full rendered prompts by default; transcripts may contain sensitive content.
-- Avoid logging prompt content unless explicit debug mode is enabled.
-- Treat generated artifacts as potentially sensitive.
-- Use session-scoped, access-controlled output paths.
-- `api_key_env` names should come from environment management, not embedded secrets.
-
-## 20. Initial D&D Artifact Generation Examples
-
-These are examples only. Use prompt IDs from the deployed prompt library.
-
-Session recap:
-
-```bash
-scriptorium run \
-  --prompt dnd.session_recap \
-  --input transcript=/work/session-42/transcript.polished.md \
-  --input glossary=/work/session-42/glossary.yml \
-  --out /work/session-42/artifacts/session_recap.md
-```
-
-Structured events:
-
-```bash
-scriptorium run \
-  --prompt dnd.structured_events \
-  --input transcript=/work/session-42/transcript.polished.md \
-  --out /work/session-42/artifacts/structured_events.json
-```
-
-Glossary suggestions:
-
-```bash
-scriptorium run \
-  --prompt dnd.glossary_suggestions \
-  --input transcript=/work/session-42/transcript.polished.md \
-  --input previous_recap=/work/session-41/artifacts/session_recap.md \
-  --out /work/session-42/artifacts/glossary_suggestions.md
-```
-
-Player-facing summary:
-
-```bash
-scriptorium run \
-  --prompt dnd.player_summary \
-  --input transcript=/work/session-42/transcript.polished.md \
-  --input structured_events=/work/session-42/artifacts/structured_events.json \
-  --out /work/session-42/artifacts/player_summary.md
-```
-
-## 21. Non-Goals
-
-Initial Narratio integration should not:
-
-- call Scriptorium internal Go packages
-- use HTTP API as the primary path
-- expect Scriptorium to read S3 refs directly
-- make Scriptorium responsible for Narratio stage state
-- make Scriptorium responsible for notification
-- require Scriptorium to understand D&D workflow semantics beyond prompt definitions
-
-## 22. Future Extension Notes
-
-Possible later extensions:
-
-- HTTP API integration
-- S3 artifact references if Scriptorium adds S3 reader support
-- richer render diagnostics and policy controls
-- token budgeting/prompt-size checks
-- batch execution if Scriptorium later adds batch support
+The request timeout and parent cancellation bound each invocation. Internal
+runner composition is documented in
+[the adapter implementation guide](../internal/adapters.md).
+
+## Request Contract
+Both request types carry:
+- binary/config/prompt/profile IDs;
+- input map and vars map;
+- output path;
+- timeout;
+- generated config + stdout/stderr log paths;
+- optional API-key env var name;
+- optional working directory.
+
+## Result Contract
+`ArtifactResult` returns:
+- output/log/generated-config paths;
+- exit code and duration;
+- command mode (`run` or `render`);
+- prompt/profile provenance;
+- `ValidationFailed` marker;
+- metadata map.
+
+## Validation and Failure Semantics
+Request validation fails for:
+- missing binary, prompt id, or output path;
+- non-positive timeout;
+- empty input/var names;
+- empty input path values;
+- missing required credential env var when `APIKeyEnv` is set.
+
+Run behavior:
+- subprocess errors propagate with context;
+- `run` exit code `2` is mapped to `ValidationFailed=true`;
+- successful subprocess still fails if output file is missing or empty.
+
+Each artifact result is limited to 64 MiB and must be a regular file without
+symlinked path components.
+
+Render behavior:
+- subprocess errors propagate;
+- output file must exist and be non-empty.
+
+## Deterministic Behavior
+- input and var maps are sorted into deterministic `--input` and `--var` CLI args.
+- stage wiring adds `session_id=narratio-session-<session_id>` to every Scriptorium request for sticky upstream routing, overriding any configured `vars.session_id`.
+- generated invocation YAML (`scriptorium.generated.v1`) is emitted when requested.
+- adapter is stateless and does not own artifact-selection policy.
+
+## Configuration
+
+Operator-selected values are defined under `pipeline.scriptorium.*`, including
+per-artifact settings under `pipeline.scriptorium.artifacts.*`, in the
+[configuration reference](../config.md#pipeline).
+
+Maintained examples with Scriptorium config:
+
+- [Full annotated pipeline](../../examples/pipeline.full.annotated.yml)
+- [Production-shaped pipeline](../../examples/pipeline.production.yml)

docs/integrations/seriatim.md

@@ -1,403 +1,62 @@
-# seriatim
+# Integration: Seriatim
 
-`seriatim` merges per-speaker WhisperX-style JSON transcripts into a single JSON transcript that preserves speaker identity and chronological order.
+## Purpose
+Define the Seriatim adapter contract used by `merge`, `normalize`, `trim`, and `render`.
 
-The current implementation supports the `merge` command. It reads one or more input JSON files, optionally maps each input file to a canonical speaker using `speakers.yml`, sorts all segments by timestamp, detects and resolves overlaps when word-level timing is available, assigns consecutive numeric `id` values, and writes a merged JSON artifact.
+## External Boundary
 
-## Usage
+Narratio invokes Seriatim as a subprocess in these modes:
 
-Run from source:
+- `seriatim merge`
+- `seriatim normalize`
+- `seriatim trim`
+- `seriatim render`
 
-```sh
-go run ./cmd/seriatim merge \
-  --input-file samples/raw/2026-04-19-Eric_Rakestraw.json \
-  --input-file samples/raw/2026-04-19-Mike_Brown.json \
-  --output-file merged.json
-```
+The configured timeout and parent cancellation bound each invocation. Internal
+runner composition is documented in
+[the adapter implementation guide](../internal/adapters.md).
 
-Optional report output:
+## Request/Result Contracts
+- `MergeRequest`/`MergeResult`: multi-input merge to base transcript, optional report.
+- `NormalizeRequest`/`NormalizeResult`: transcript normalization with explicit schema.
+- `TrimRequest`/`TrimResult`: transcript trimming with required keep selector.
+- `RenderRequest`/`RenderResult`: transcript-to-markdown rendering with explicit format and render booleans.
 
-```sh
-go run ./cmd/seriatim merge \
-  --input-file eric.json \
-  --input-file mike.json \
-  --output-file merged.json \
-  --report-file report.json
-```
+Results include output/log/config paths, timing, exit code, and metadata.
 
-## CLI
+## Validation and Failure Semantics
+Runner construction validates:
+- binary presence;
+- timeout > 0;
+- supported output schema (`seriatim-minimal|seriatim-intermediate|seriatim-full`);
+- non-negative coalesce gap.
 
-```text
-seriatim merge [flags]
-```
+Invocation fails on:
+- missing required request paths/inputs;
+- invalid normalize schema override;
+- unsupported render format;
+- subprocess failure;
+- invalid JSON outputs for merge/normalize/trim;
+- missing `segments` array for normalize/trim transcript outputs;
+- empty render output files.
 
-Global flags:
+When report paths are provided/enabled, report files must parse as JSON.
+Each Seriatim JSON or rendered-text result is limited to 64 MiB and must be a
+regular file without symlinked path components.
 
-| Flag | Description |
-| --- | --- |
-| `--help` | Show command help. |
-| `--version` | Show application version. Local builds default to `dev`; release builds inject the release version. |
+## Deterministic Behavior
+- argument ordering is deterministic per command construction.
+- merge env overrides are explicit (`SERIATIM_*`) and only emitted when configured.
+- generated invocation YAML (`seriatim.generated.v1`) is emitted when requested.
+- adapter does not write manifests or choose stage inputs.
 
-`merge` flags:
+## Configuration
 
-| Flag | Required | Default | Description |
-| --- | --- | --- | --- |
-| `--input-file` | Yes | none | Input transcript JSON file. Repeat once per speaker/input file. |
-| `--output-file` | Yes | none | Merged transcript JSON output path. |
-| `--report-file` | No | none | Optional report JSON output path. |
-| `--speakers` | No | none | Speaker map YAML file. When omitted, input file basenames are used as speaker labels. |
-| `--autocorrect` | No | none | Autocorrect rules YAML file. When omitted, the default `autocorrect` module leaves text unchanged. |
-| `--input-reader` | No | `json-files` | Input reader module. |
-| `--output-modules` | No | `json` | Comma-separated output modules. |
-| `--output-schema` | No | `seriatim-intermediate` | JSON output contract. Allowed values are `seriatim-minimal`, `seriatim-intermediate`, and `seriatim-full`. If omitted, the runtime default is used; consumers that depend on a specific shape should set this explicitly. |
-| `--preprocessing-modules` | No | `validate-raw,normalize-speakers,trim-text` | Comma-separated preprocessing modules, evaluated in order. |
-| `--postprocessing-modules` | No | `detect-overlaps,resolve-overlaps,backchannel,filler,resolve-danglers,coalesce,detect-overlaps,autocorrect,assign-ids,validate-output` | Comma-separated postprocessing modules, evaluated in order. |
-| `--coalesce-gap` | No | `3.0` | Maximum same-speaker gap in seconds for `coalesce`; also used as the `resolve-overlaps` context window. Must be a non-negative float. |
+Operator-selected values are defined under `pipeline.seriatim.*` and
+`pipeline.render.*` in the
+[configuration reference](../config.md#pipeline).
 
-Environment variables:
+Maintained examples with Seriatim config:
 
-| Environment Variable | Default | Description |
-| --- | --- | --- |
-| `SERIATIM_OUTPUT_SCHEMA` | `seriatim-intermediate` | Output schema used when `--output-schema` is not explicitly provided. Allowed values are `seriatim-minimal`, `seriatim-intermediate`, and `seriatim-full`. The CLI flag takes precedence. |
-| `SERIATIM_OVERLAP_WORD_RUN_GAP` | `1.0` | Maximum gap in seconds between adjacent timed words when `resolve-overlaps` builds word-run replacement segments. Must be a positive float. |
-| `SERIATIM_OVERLAP_WORD_RUN_REORDER_WINDOW` | `1.0` | Near-start window in seconds for ordering replacement word runs shortest-first. Must be a positive float. |
-| `SERIATIM_BACKCHANNEL_MAX_DURATION` | `2.0` | Maximum duration in seconds for `backchannel` classification. Must be a positive float. |
-| `SERIATIM_FILLER_MAX_DURATION` | `1.25` | Maximum duration in seconds for `filler` classification. Must be a positive float. |
-
-## Input JSON Format
-
-Each input file must be valid JSON with a top-level `segments` array. The current parser accepts the WhisperX segment subset needed for merging:
-
-```json
-{
-  "segments": [
-    {
-      "start": 1.25,
-      "end": 3.5,
-      "text": "Hello there.",
-      "words": [
-        {"word": "Hello", "start": 1.25, "end": 1.55, "score": 0.98},
-        {"word": "there.", "start": 1.7, "end": 2.0}
-      ]
-    }
-  ]
-}
-```
-
-Required segment fields:
-
-- `start`: number, must be `>= 0`.
-- `end`: number, must be `>= start`.
-- `text`: string.
-
-Optional word fields:
-
-- `words`: array of word timing objects.
-- `words[].word`: string.
-- `words[].start`: optional number, must be `>= 0` when present.
-- `words[].end`: optional number, must be `>= start` when present with `start`.
-- `words[].score`: optional number.
-- `words[].speaker`: optional raw speaker label string.
-
-Word-level timing is preserved internally for overlap resolution. If a word is missing `start` or `end`, seriatim keeps the word text, emits a warning in the optional report, and does not use that word as a timing anchor. Word timing is not emitted in the final JSON artifact.
-
-## Speaker Map Format
-
-`speakers.yml` maps input files to canonical speaker names using ordered substring rules:
-
-This file is optional. If `--speakers` is omitted, `seriatim` uses each input file basename as the segment speaker label.
-
-```yaml
-match:
-  - speaker: "Eric Rakestraw"
-    match:
-      - "Eric_Rakestraw"
-      - "Eric"
-
-  - speaker: "Mike Brown"
-    match:
-      - "Mike_Brown"
-      - "mb"
-```
-
-For each `--input-file`, `seriatim` takes the file basename and evaluates the rules in order. The first rule with a matching substring wins, and no later rules are evaluated.
-
-For example, this input:
-
-```text
-samples/raw/2026-04-19-Eric_Rakestraw.json
-```
-
-matches this rule because the basename contains `Eric_Rakestraw`:
-
-```yaml
-- speaker: "Eric Rakestraw"
-  match:
-    - "Eric_Rakestraw"
-```
-
-Important details:
-
-- Matching is against the input file basename, not the full path.
-- Matching is case-insensitive.
-- Rules are evaluated from first to last.
-- Each rule must have a non-empty `speaker`.
-- Each rule must have at least one non-empty `match` string.
-- Duplicate speaker names are invalid.
-- Every input file must match at least one rule or the command fails.
-
-Deprecated old format:
-
-```yaml
-inputs:
-  eric.json:
-    speaker: "Eric Rakestraw"
-```
-
-The old `inputs:` direct mapping format is no longer supported.
-
-## Output JSON Format
-
-`--output-modules json` controls the writer. `--output-schema` controls the JSON contract that writer serializes.
-
-The named schemas are stable public contracts. If a consumer depends on a specific shape, it should request that schema explicitly at runtime. The runtime default selection may change in a future release.
-
-The `seriatim-intermediate` schema is the current default selection when neither `--output-schema` nor `SERIATIM_OUTPUT_SCHEMA` is set. It stays close to the minimal schema, but adds optional `categories` on each segment:
-
-```json
-{
-  "metadata": {
-    "application": "seriatim",
-    "version": "dev",
-    "output_schema": "seriatim-intermediate"
-  },
-  "segments": [
-    {
-      "id": 1,
-      "start": 1.25,
-      "end": 3.5,
-      "speaker": "Eric Rakestraw",
-      "text": "Hello there.",
-      "categories": ["backchannel"]
-    }
-  ]
-}
-```
-
-The `seriatim-full` schema uses the full seriatim envelope:
-
-```json
-{
-  "metadata": {
-    "application": "seriatim",
-    "version": "dev",
-    "input_reader": "json-files",
-    "input_files": ["eric.json", "mike.json"],
-    "preprocessing_modules": ["validate-raw", "normalize-speakers", "trim-text"],
-    "postprocessing_modules": ["detect-overlaps", "resolve-overlaps", "backchannel", "filler", "resolve-danglers", "coalesce", "detect-overlaps", "autocorrect", "assign-ids", "validate-output"],
-    "output_modules": ["json"]
-  },
-  "segments": [
-    {
-      "id": 1,
-      "source": "eric.json",
-      "source_segment_index": 0,
-      "speaker": "Eric Rakestraw",
-      "start": 1.25,
-      "end": 3.5,
-      "text": "Hello there.",
-      "overlap_group_id": 1
-    },
-    {
-      "id": 2,
-      "source": "eric.json",
-      "source_ref": "word-run:1:1:1",
-      "derived_from": ["eric.json#0"],
-      "speaker": "Eric Rakestraw",
-      "start": 2.0,
-      "end": 2.5,
-      "text": "Resolved word run",
-      "categories": ["backchannel"]
-    }
-  ],
-  "overlap_groups": [
-    {
-      "id": 1,
-      "start": 1.25,
-      "end": 4.0,
-      "segments": ["eric.json#0", "mike.json#0"],
-      "speakers": ["Eric Rakestraw", "Mike Brown"],
-      "class": "unknown",
-      "resolution": "unresolved"
-    }
-  ]
-}
-```
-
-The `seriatim-minimal` schema emits minimal metadata and compact ordered segments:
-
-```json
-{
-  "metadata": {
-    "application": "seriatim",
-    "version": "dev",
-    "output_schema": "seriatim-minimal"
-  },
-  "segments": [
-    {
-      "id": 1,
-      "start": 1.25,
-      "end": 3.5,
-      "speaker": "Eric Rakestraw",
-      "text": "Hello there."
-    }
-  ]
-}
-```
-
-Minimal output intentionally omits categories, overlap groups, source/provenance fields, and pipeline configuration metadata.
-
-Intermediate output intentionally omits overlap groups and source/provenance fields, but keeps optional `categories` and minimal metadata.
-
-Segments are sorted deterministically by:
-
-```text
-(start, end, source, source_segment_index/source_ref, speaker)
-```
-
-Final segment IDs are assigned after sorting and start at `1`.
-
-The public Go output contract is available from:
-
-```go
-import "gitea.maximumdirect.net/eric/seriatim/schema"
-```
-
-The same package embeds machine-readable JSON Schemas in `schema/full-output.schema.json`, `schema/intermediate-output.schema.json`, and `schema/minimal-output.schema.json`. The default `validate-output` postprocessor validates the selected output shape and verifies final segment IDs are present, sequential, and start at `1`.
-
-## Overlap Detection
-
-The default postprocessing pipeline detects overlapping segment groups.
-
-Overlap behavior:
-
-- A strict timing overlap is required: `next.start < current_group_end`.
-- Segments that only touch at a boundary are not grouped.
-- Groups require at least two distinct speakers.
-- Transitive overlaps are grouped together.
-- Segments in detected groups receive `overlap_group_id`.
-- `overlap_groups[].segments` contains stable references in `source#source_segment_index` format.
-- `class` is currently `unknown`.
-- `resolution` is `unresolved` until `resolve-overlaps` replaces the group.
-
-## Overlap Resolution
-
-The default postprocessing pipeline runs `detect-overlaps`, then `resolve-overlaps`, then `backchannel`, then `filler`, then `resolve-danglers`, then `coalesce`, then a second `detect-overlaps` pass.
-
-For each detected overlap group, `resolve-overlaps` uses preserved WhisperX word timing to build smaller word-run replacement segments:
-
-- The resolution window expands the detected overlap group by `--coalesce-gap` seconds on both sides.
-- Nearby same-speaker context segments are included when they intersect the expanded window and their start or end is within `--coalesce-gap` of the original overlap boundary.
-- Once a segment is selected for replacement, all timed words from that segment participate in word-run construction; the window controls segment selection, not per-word clipping.
-- Context segments that are part of another detected overlap group are not pulled into the current group.
-- Untimed words are included in replacement text in original word order when nearby timed words create a replacement run.
-- Untimed words do not affect replacement segment start/end times or word-run gap splitting.
-- Words for the same speaker are merged into one run when the gap between adjacent words is no greater than `SERIATIM_OVERLAP_WORD_RUN_GAP`.
-- The default word-run gap is `1.0` seconds.
-- Set `SERIATIM_OVERLAP_WORD_RUN_GAP` to a positive number of seconds to override the default.
-- Near-start replacement word runs are reordered so shorter segments come first when adjacent starts are within `SERIATIM_OVERLAP_WORD_RUN_REORDER_WINDOW`.
-- The default word-run reorder window is `1.0` seconds.
-- Set `SERIATIM_OVERLAP_WORD_RUN_REORDER_WINDOW` to a positive number of seconds to override the default.
-- Replacement segment text is built by joining word text with single spaces.
-- Replacement segments include `source_ref` and `derived_from`.
-- Replacement segments omit `source_segment_index` because they are derived from one or more original segments.
-- Resolved overlap groups are removed before the second detection pass.
-- Replacement segments are left without `overlap_group_id` until the second detection pass annotates any remaining overlap.
-- If a speaker has no usable word timing in a group, that speaker's original segment is kept.
-- If no speakers in a group have usable word timing, the original group and annotations remain unchanged.
-
-## Backchannels
-
-The default pipeline runs `backchannel` before `coalesce`. It tags short acknowledgement segments with:
-
-```json
-"categories": ["backchannel"]
-```
-
-Backchannel matching is case-insensitive, ignores punctuation for matching and word-count purposes, trims surrounding whitespace, and requires a matching acknowledgement phrase, no more than three whitespace-delimited words, and duration no greater than `SERIATIM_BACKCHANNEL_MAX_DURATION` seconds. The default maximum duration is `2.0` seconds.
-
-## Fillers
-
-The default pipeline runs `filler` after `backchannel` and before `coalesce`. It tags short filler utterances with:
-
-```json
-"categories": ["filler"]
-```
-
-Filler matching is case-insensitive, ignores punctuation for matching and word-count purposes, trims surrounding whitespace, and requires only filler tokens such as `um`, `uh`, `er`, `erm`, `ah`, `eh`, `hmm`, `mm`, or repeated combinations of those tokens. Matching segments must contain no more than three whitespace-delimited words and have duration no greater than `SERIATIM_FILLER_MAX_DURATION` seconds. The default maximum duration is `1.25` seconds.
-
-## Dangler Resolution
-
-The default pipeline runs `resolve-danglers` before `coalesce` and before the second overlap detection pass. It repairs short derived fragments when they share provenance with a nearby segment:
-
-- Dangling-end fragments have no more than two words and end in punctuation.
-- Dangling-start fragments have no more than two words.
-- Matching uses same-speaker segments with any shared `derived_from` value.
-- Merged segments use `source_ref` values such as `resolve-danglers:1`, keep the target segment's transcript position, and union `derived_from`.
-
-## Coalescing
-
-The default pipeline runs `coalesce` after `resolve-danglers` and before the second overlap detection pass. It merges adjacent same-speaker segments in the transcript's current order when `next.start - current.end <= --coalesce-gap`.
-
-Coalesced segments use `source_ref` values such as `coalesce:1`, include `derived_from`, and omit `source_segment_index`.
-
-Different-speaker backchannel and filler segments do not block coalescing of surrounding same-speaker segments. Same-speaker backchannel and filler segments are merged normally when they are within `--coalesce-gap`. When same-speaker segments are coalesced, any `backchannel` or `filler` category from the merged inputs is dropped from the coalesced segment.
-
-## Autocorrect
-
-Autocorrect is included in the default postprocessing pipeline. If `--autocorrect` is omitted, the module leaves transcript text unchanged and records a skip event in the optional report.
-
-Enable corrections by passing `--autocorrect`:
-
-```sh
-go run ./cmd/seriatim merge \
-  --input-file input.json \
-  --autocorrect autocorrect.yml \
-  --output-file merged.json
-```
-
-`autocorrect.yml` format:
-
-```yaml
-autocorrect:
-  - target: "Hrank"
-    match:
-      - "hrank"
-      - "Frank"
-
-  - target: "Mike Brown"
-    match:
-      - "Mike Pat"
-```
-
-Matching behavior:
-
-- Matching is case-sensitive.
-- Matches apply only to whole tokens, not substrings inside larger words.
-- Punctuation and whitespace can surround a match.
-- Multi-word and hyphenated matches are supported.
-- Duplicate match strings are invalid, including duplicates across separate rules.
-
-## Current Limitations
-
-- Only JSON input is supported.
-- Overlap resolution depends on WhisperX word timing; groups without usable word timing remain unresolved.
-- Alternate output formats are not implemented yet.
-
-## Release Builds
-
-Local builds record version metadata as `dev`. Release builds should inject the release version with `ldflags`:
-
-```sh
-go build -ldflags "-X gitea.maximumdirect.net/eric/seriatim/internal/buildinfo.Version=v1.0.0" ./cmd/seriatim
-```
+- [Full annotated pipeline](../../examples/pipeline.full.annotated.yml)
+- [Production-shaped pipeline](../../examples/pipeline.production.yml)

docs/integrations/whisperx.md

@@ -0,0 +1,71 @@
+# Integration: WhisperX
+
+## Purpose
+
+WhisperX transcribes each prepared speaker audio file for Narratio's
+`transcribe` stage. Narratio uses an HTTP boundary and installs each successful
+response as that speaker's raw transcript JSON.
+
+## HTTP Boundary
+
+Narratio sends an HTTP `POST` to the configured transcription URL using
+`multipart/form-data` with:
+
+- `file`: the audio file, retaining its base filename; and
+- `language`: the configured language string.
+
+The server must return a `2xx` response whose body is valid JSON. Narratio does
+not currently require a more specific response schema at this boundary.
+
+The transcription URL must be an absolute `http` or `https` URL. The audio body
+is streamed through a fresh multipart writer for every attempt, so its memory
+use is bounded by the transport buffer rather than by the complete audio file.
+WhisperX response acquisition is capped at 10 MiB.
+
+## Request And Result Contract
+
+Each adapter request identifies a speaker, a readable audio file, and the
+destination for the raw transcript. The HTTP request carries the audio and
+language; the speaker identifier remains Narratio orchestration metadata.
+
+On success, Narratio atomically writes the response body to the requested
+destination. The adapter result reports that logical output together with the
+attempt count, final HTTP status when available, elapsed duration, and adapter
+identity metadata. A failed or invalid response is not installed as the
+transcript output.
+
+## Retry, Timeout, And Cancellation
+
+- The configured timeout applies independently to each HTTP attempt.
+- `retries` means additional attempts after the first.
+- HTTP `429`, HTTP `5xx`, attempt timeouts, and network errors are retryable.
+- Other HTTP `4xx` responses and explicit cancellation are not retryable.
+- Narratio waits the configured retry delay between attempts and aborts that
+  wait when the parent context is canceled.
+
+## Validation And Failure Semantics
+
+Client construction rejects a missing or non-HTTP(S) absolute transcription URL,
+a missing language, a non-positive timeout, negative retries, or a negative
+retry delay. A request fails before transmission when its audio or output path
+is missing.
+
+Non-`2xx` status, transport failure, response-size overflow, invalid JSON, or
+failure to install the output causes the transcription to fail. Errors include
+attempt context, and the result retains attempts, final status when available,
+and elapsed duration for diagnostics.
+
+## Determinism And Concurrency
+
+Each audio request has stable multipart field names, and successful bytes are
+installed atomically. The transcribe stage may process speaker files in
+parallel, bounded by the configured concurrency. It records results in stable
+speaker order after all work completes; any speaker failure fails the stage.
+
+## Related Canonical Docs
+
+- [Configuration](../config.md#pipeline) defines the operator-selected
+  WhisperX URL, language, timeouts, retry policy, and concurrency.
+- [Adapter implementation](../internal/adapters.md) describes internal wiring.
+- [Transcribe stage](../internal/stage-transcribe.md) describes stage mechanics,
+  durable artifacts, and manifests.

docs/internal/adapters.md

@@ -0,0 +1,92 @@
+# Internal: Adapters
+
+## Purpose
+
+Explain the adapter interfaces and production composition used by application
+and stage orchestration. Externally observable protocols and formats belong in
+the [integration contracts](../integrations/).
+
+## Adapter Boundaries
+
+Narratio stage logic depends on adapter interfaces, not transport-specific details.
+
+Primary adapters:
+
+- `whisperx.Client`
+- `seriatim.Runner`
+- `audita.Runner`
+- `scriptorium.Runner`
+- `notarius.Runner`
+- `storage.ObjectStore`
+- `notify.Sender`
+
+## Ownership
+
+Adapters own:
+
+- HTTP/subprocess/SDK argument and transport details.
+- Backend-specific request/response mapping.
+
+Adapters do not own:
+
+- stage ordering/skip/force logic;
+- manifest transitions;
+- canonical path policy.
+
+## Default Wiring
+
+`internal/app/runner.go` initializes default adapters when not injected:
+
+- WhisperX HTTP client from pipeline config.
+- Seriatim subprocess runner.
+- Audita subprocess runner.
+- Scriptorium subprocess runner.
+- Notarius subprocess runner when extraction is enabled.
+- Noop notifier (`notify.NoopSender`).
+- Object store only when required by selected stages/config.
+
+Notarius is composed only when extraction is enabled; the extract stage owns
+prepared reference resolution, receipt, bundle, and configured-lane policy.
+The adapter validates the ordered selector/absolute-path pairs and is the sole
+owner of serializing them as repeated `--reference` arguments before `--json`.
+
+Object-store construction goes through `newCommandObjectStore`, which loads
+configured filesystem secrets before adapter initialization.
+
+## Failure Semantics
+
+- Constructor errors fail stage execution setup early.
+- Runtime adapter errors propagate to stage code and then manifest failure handling.
+- Subprocess adapters persist stage logs/generated configs through stage-managed paths.
+- Shared subprocess execution starts an owned process group on Linux/macOS or a
+  kill-on-close job object on Windows. Every terminal path disposes of that
+  owned tree before returning. After a natural leader exit, Unix checks for
+  remaining group members and uses bounded graceful then forceful termination;
+  Windows closes the job so kill-on-close applies. Cancellation, deadlines, and
+  diagnostic limits use the same terminal disposal path without losing their
+  original result classification. Child environments contain only the execution
+  baseline and adapter-specified values; configured credentials are explicit
+  sensitive values. Stdout and stderr are redacted while streaming into separate
+  8 MiB diagnostic captures; a bounded wait closes a stream retained by a
+  departed leader's descendant. Unsupported platforms reject owned command
+  execution.
+
+## Implementation And Tests
+
+- Composition: `internal/app/runner.go`, `internal/app/object_store.go`
+- Shared subprocess mechanics: `internal/adapters/subprocess`
+- Focused adapters: `internal/adapters/{whisperx,seriatim,audita,scriptorium,notarius,storage,notify}`
+- `internal/adapters/whisperx/http_test.go`
+- `internal/adapters/seriatim/subprocess_test.go`
+- `internal/adapters/audita/subprocess_test.go`
+- `internal/adapters/scriptorium/subprocess_test.go`
+- `internal/adapters/notarius/subprocess_test.go`
+- `internal/adapters/storage/*_test.go`
+- `internal/app/runner_test.go`
+
+See the [WhisperX](../integrations/whisperx.md),
+[Seriatim](../integrations/seriatim.md), [Audita](../integrations/audita.md),
+[Scriptorium](../integrations/scriptorium.md), and
+[Notarius](../integrations/notarius.md) contracts before changing an
+externally visible boundary. Operator-selected values belong in
+[Configuration](../config.md).

docs/internal/artifacts.md

@@ -0,0 +1,229 @@
+# Internal: Artifacts
+
+## Purpose
+
+Explain the artifact registry, runtime catalog, resolver, previous-input
+requirements, and shared remote current-state mechanics implemented by
+`internal/artifacts`. Configuration fields that accept source IDs belong in
+[Configuration](../config.md); physical placement belongs in
+[Operations](../operations.md).
+
+## Built-in Source IDs
+
+The internal registry recognizes these stable built-in source IDs:
+
+- `narratio.transcript.base`
+- `narratio.transcript.polished`
+- `narratio.transcript.final`
+- `narratio.transcript.final_trimmed`
+- `narratio.transcript.final_markdown`
+- `narratio.transcript.final_trimmed_markdown`
+- `narratio.bounds.session`
+
+Registry entries bind each ID to its producer, output kind, canonical fallback,
+and content validator. The focused stage documents own their input/output flow;
+[Configuration](../config.md) owns where operators may select these IDs.
+
+## Configured, Extraction, And Previous-Session Sources
+
+- configured source ID format: `narratio.artifact.<artifact_key>`
+- extraction source ID format: `narratio.extraction.<output_key>`
+- previous-session source ID format: `narratio.previous_session.artifact.<artifact_key>`
+
+All formats are validated by strict source-policy rules. Configured artifact and
+extraction keys use `^[a-z][a-z0-9_]*$`; source parsers never normalize an
+unrecognized token into a valid source. Extraction sources are registered only
+from `pipeline.notarius.outputs`; the Notarius index has no selectable source
+ID.
+
+Prepared stable source IDs are `narratio.input.players`,
+`narratio.input.party`, `narratio.input.glossary`, and
+`narratio.input.spell_catalog`. Artifact policy owns their canonical manifest
+kind and prepared filename vocabulary.
+
+## Runtime Catalog
+
+`ArtifactCatalog` tracks:
+
+- `planned`: source registered for run context;
+- `executable`: included in the effective analyze artifact set;
+- `available`: local file exists and validates;
+- `provenance`: availability source.
+
+Configured definitions are always registered. Without an explicit selection,
+the effective analyze set contains enabled definitions. With `--artifacts`, the
+exact named configured definitions become the effective set for that invocation,
+regardless of their `enabled` value; dependencies are not added implicitly.
+Availability is separate from executability: a non-executable configured output
+may be reused from a canonical non-empty file, while an executable definition
+is generated by analyze. Extraction entries are registered from configuration
+and become available only after compatible extraction evidence is hydrated.
+
+Current provenance values:
+
+- `generated.current_analyze_run`
+- `filesystem.disabled_artifact_output`
+- `manifest.inputs.previous_cache`
+- `current_session.previous_cache`
+
+## Resolution Rules
+
+Built-ins:
+
+1. manifest producer outputs (when present)
+2. canonical session-path fallback
+
+Configured sources (`narratio.artifact.*`):
+
+- resolve only through runtime catalog availability.
+
+Prepared stable sources (`narratio.input.*`):
+
+- resolve only from the current manifest's exact prepared-input record;
+- require the policy-owned canonical path below the session root, a confined
+  non-symlink regular file, a non-empty payload, and a matching SHA-256
+  checksum; and
+- return an immutable source/path/checksum/size identity shared by extract and
+  analyze rather than falling back to campaign/session source paths.
+
+Extraction sources (`narratio.extraction.*`):
+
+- use the shared typed bundle evidence inspection in `extraction_evidence.go`;
+- require a current successful extract record with the exact configured source,
+  compatible contract and Notarius provenance, a confined regular durable
+  payload, matching checksum, and the current resolved trimmed-transcript
+  identity;
+- remain unavailable unless catalog hydration receives valid evidence. Resume
+  treats absent or obsolete evidence as a rerun decision and unsafe evidence as
+  an error; and
+- are never inferred by scanning the Notarius bundle directory.
+
+Previous-session sources (`narratio.previous_session.artifact.*`):
+
+- resolve only from local `previous/` cache state;
+- prefer manifest-backed previous-input paths;
+- fallback to existing previous-cache filesystem paths.
+
+Source absence is evaluated by the consuming artifact input. An optional input
+is omitted from that invocation; a required input fails resolution. This is
+separate from a stage's lifecycle outcome.
+
+Validation by content type:
+
+- transcript JSON built-ins: JSON with top-level `segments` array;
+- transcript Markdown built-ins: non-empty text file;
+- bounds built-in: valid JSON;
+- configured/previous-session artifact files: non-empty text file.
+
+## Previous Requirement Collection
+
+`CollectPreviousArtifactRequirements`:
+
+- scans the effective configured artifact set;
+- extracts only canonical previous-session sources;
+- deduplicates by artifact key;
+- merges required and optional references (required wins);
+- returns deterministic ordering and source locations.
+
+## Current-State Helpers
+
+Artifacts package owns shared remote current-state loading mechanics used by
+restore, status and validation checks, and previous-cache planning.
+
+For a new-protocol current state, the pointer-selected immutable commit is the
+complete restore authority. Callers receive its declared object identities and
+must not supplement them by listing mutable session prefixes. The legacy reader
+is intentionally separate and remains migration-only support.
+
+The reader opens each small control object directly and enforces owner-specific
+limits before decoding: 64 KiB for the mutable commit pointer, 4 MiB for the
+immutable commit manifest, and 8 MiB for the selected session manifest. Legacy
+compatibility applies a 4 KiB limit to `current/run_id.txt` and the same 8 MiB
+manifest limit to `current/manifest.json`. These are exposed as
+`MaxCurrentCommitPointerBytes`, `MaxRemoteCommitManifestBytes`,
+`MaxRemoteSessionManifestBytes`, `MaxLegacyCurrentRunPointerBytes`, and
+`MaxLegacyCurrentManifestBytes`.
+
+Each read uses the generation and size metadata returned with its opened body.
+Actual bytes remain subject to a limit-plus-one read even if size metadata is
+absent or inaccurate. Immutable selections then retain their declared-size,
+checksum, generation, and identity checks. No current-state control object is
+downloaded through a temporary file.
+
+Core helpers:
+
+- `LoadCurrentState`
+- `ValidateCurrentStateIdentity`
+- `RemoteCommitManifest` and `CurrentCommitPointer`
+
+Typed missing-state errors:
+
+- `CurrentRunPointerMissingError` (`ErrCurrentRunPointerMissing`)
+- `CurrentManifestMissingError` (`ErrCurrentManifestMissing`)
+
+Identity validation supports caller-provided expectations:
+
+- expected campaign;
+- expected session ID;
+- expected run ID, or pointer/manifest run-ID consistency check.
+
+Caller policy is intentionally outside artifacts helpers:
+
+- some callers fail on missing current state;
+- some callers downgrade missing state to status/findings;
+- some callers skip optional behavior when state is missing.
+
+## Key Path Helpers
+
+`internal/artifacts/paths.go` and S3-key helpers define canonical helpers for:
+
+- session/work/run paths;
+- previous-cache paths;
+- spool/cache paths;
+- S3 session/run/current-state key layout.
+
+New publication creates run-scoped immutable objects, including
+`runs/{run_id}/commit.json` and `runs/{run_id}/session-manifest.json`. The sole
+mutable selector is `current/commit-pointer.json`; readers verify its selected
+commit and declared object generations/checksums. Legacy current-pair loading
+is confined to `current_state_legacy.go` for migration only.
+
+Campaign, session, and Narratio run IDs are validated as portable opaque
+segments at configuration and artifact boundaries before they can be used in a
+workspace or S3 namespace. Previous-artifact destinations remain typed,
+multi-segment relative paths and are confined beneath `previous/artifacts`; they
+are not treated as opaque identifiers.
+
+See [Workspace Internals](workspace.md) for how callers consume local helpers
+and [Operations](../operations.md#local-state-layout) for the authoritative
+physical layout.
+
+## Invariants
+
+- source ID formats are stable contracts;
+- artifact resolution is deterministic and manifest-aware;
+- extraction sources are available only from a compatible successful manifest
+  record;
+- previous-session source resolution in `analyze` is local-only;
+- remote current-state key construction remains centralized in artifacts helpers.
+
+## Implementation And Tests
+
+- Registry and resolution: `internal/artifacts/artifact_resolver.go`,
+  `internal/artifacts/catalog.go`, `internal/artifacts/transcripts.go`,
+  `internal/artifacts/extraction_catalog.go`,
+  `internal/artifacts/extraction_evidence.go`,
+  `internal/artifacts/extraction_input.go`,
+  `internal/artifacts/prepared_input.go`
+- Current state: `internal/artifacts/current_state.go`,
+  `internal/artifacts/current_state_commit.go`,
+  `internal/artifacts/current_state_legacy.go`
+- Paths and keys: `internal/artifacts/paths.go`,
+  `internal/artifacts/s3_keys.go`
+- Previous requirements: `internal/artifacts/previous_requirements.go`
+- Tests: `internal/artifacts/artifact_resolver_test.go`,
+  `internal/artifacts/catalog_test.go`,
+  `internal/artifacts/extraction_catalog_test.go`,
+  `internal/artifacts/current_state_test.go`,
+  `internal/artifacts/paths_model_test.go`,
+  `internal/artifacts/previous_requirements_test.go`

docs/internal/command-restore.md

@@ -0,0 +1,110 @@
+# Internal: Command Restore
+
+## Purpose
+
+Explain the implemented restore discovery, planning, installation, and
+reporting flow in `internal/app`. User invocation belongs in
+[CLI](../cli.md#session-restore), and the operator recovery procedure and
+physical restore scope belong in
+[Operations](../operations.md#restore-workflow).
+
+Restore separates remote authority, local conflict policy, and filesystem
+mutation so each remains testable independently.
+
+## Discovery Contract
+
+Discovery delegates current-state pointer and manifest loading to
+`internal/artifacts`, then validates the result against the resolved request:
+
+- campaign must match;
+- session ID must match.
+- run ID must match the pointer-selected committed run.
+
+Restore treats any missing or invalid remote current state as a command error.
+
+## Planning Contract
+
+Restore planner action kinds:
+
+- `download`;
+- `skip_same`;
+- `conflict`.
+
+Planner behavior:
+
+- a new-protocol restore uses only the selected commit's declared artifact set;
+  each action carries that artifact's immutable key, checksum, size, and
+  generation. Coherent legacy state remains on the isolated compatibility path;
+- remote-to-local mapping is traversal-safe;
+- actions are sorted by local relative path and then remote key;
+- force converts differing eligible regular files from conflicts to downloads;
+  directories and other non-regular targets remain conflicts.
+
+For a non-dry-run restore, planning/classification happens only after acquiring
+the session lock. Runner manifest/reuse checks acquire that same lock first.
+
+Previous-cache readiness is resolved through `previouscache.Resolve` for restore,
+prepare, status, and validation. A committed source is selected only by its
+exact source identity; legacy fallback remains isolated and rejects ambiguity.
+
+## Execution Contract
+
+Execution order and safety:
+
+- non-manifest downloads happen before manifest install;
+- `manifest.json` installs last;
+- downloads use sibling temp files plus atomic rename;
+- manifest replacement is validated before rename;
+- each committed object is verified against its declared checksum, size, and
+  generation before installation;
+- a committed manifest already verified during discovery is retained for the
+  matching restore action and revalidated before installation, avoiding a
+  second body transfer;
+- failed installs do not roll back files already written in the same execution.
+- a durable `.restore-incomplete.json` marker is written before installation.
+  It blocks runners until a restore retry completes all verified installs and
+  the local manifest replacement, at which point it is removed.
+- restored manifest local references are rebased beneath the selected local
+  session root. Unsafe relative references and producer-machine absolute paths
+  outside the manifest's producer session root are rejected; producer-local
+  spool/cache and cleanup locations are not restored as authority.
+
+Audio restore path:
+
+- uses `audio.MaterializeS3Audio`;
+- integrates spool and S3 audio cache paths;
+- reuses cached audio only when its no-follow regular file, content digest, and
+  identity sidecar all match the selected remote object version; otherwise it
+  refreshes through the durable download path.
+
+## Reporting Contract
+
+- dry-run mode prints a summary, performs no durable session writes, and may
+  read remote current-state or object-identity data to produce that summary;
+- execution mode persists the canonical restore report described in
+  [Operations](../operations.md#restore-workflow);
+- report includes plan counts, per-action status, and execution failures.
+
+## Invariants
+
+- restore uses committed remote current state as authority;
+- one restore or status inspection observes the single pointer-selected commit
+  loaded at discovery; later pointer changes cannot add objects or substitute a
+  different run into its plan;
+- a verified `current/commit-pointer.json` and its selected immutable commit
+  establish new-protocol remote commitment; coherent legacy
+  `current/run_id.txt` plus `current/manifest.json` remains read-only migration
+  support;
+- restore does not execute pipeline stages.
+
+## Implementation And Tests
+
+- Discovery: `internal/app/restore_discovery.go`
+- Planning: `internal/app/restore_plan.go`, `internal/previouscache`
+- Execution: `internal/app/restore_execute.go`
+- Reporting and command coordination: `internal/app/restore_report.go`,
+  `internal/app/restore.go`
+- Tests: `internal/app/restore_discovery_test.go`,
+  `internal/app/restore_plan_test.go`,
+  `internal/app/restore_execution_test.go`,
+  `internal/app/restore_workflow_test.go`

docs/internal/fileops.md

@@ -0,0 +1,71 @@
+# Internal: File Operations
+
+`internal/fileops` owns the narrow mechanics for durable replacement of one
+byte file. Callers keep ownership of serialization, validation, cancellation,
+and destination-directory policy.
+
+## Destination Confinement
+
+Before it creates, replaces, or installs a destination file, `fileops` opens
+each ancestor from the filesystem root and rejects symbolic links or components
+that change during traversal. The resulting parent-directory handle is retained
+for sibling temporary-file creation and rename, so a later pathname swap cannot
+redirect the replacement. Existing destination symlinks are replaced as leaf
+entries; their targets are never followed.
+
+Remote object acquisition uses a writer supplied by the storage owner. The
+writer receives a `fileops`-owned, already-open sibling temporary file rather
+than a mutable destination path. Callers still own remote object selection,
+validation, conflict handling, and final mode.
+
+Directory promotion keeps the verified destination parent open while it creates
+the temporary tree, copies regular source entries, and performs the platform
+no-replace rename. Platforms without a verified handle-relative atomic
+no-replace primitive reject promotion before writing a temporary tree.
+
+## Cleanup Contract
+
+`RemoveAllUnderRoot` accepts an explicit root and a proper descendant. It opens
+the root and each target ancestor without following symlinks, then removes the
+tree through those directory handles. It rejects root deletion and any symlink
+encountered in the target path or tree; repeated removal of a missing target is
+successful. Command and post-publish policy remains owned by `internal/app`.
+
+## Confined Reads
+
+`ReadRegularFileUnderRoot` is the no-follow, bounded read primitive for a
+caller-selected root and relative file path; `ReadRegularFile` is its
+path-based convenience wrapper. They verify every ancestor through directory
+handles and admit only a stable regular-file handle. Callers enforce their own
+byte limits and access policy. Credential mode policy and environment
+precedence remain owned by `internal/app`.
+
+## Replacement Contract
+
+`ReplaceFileAtomic` requires an existing destination directory. It creates a
+sibling temporary file, writes the complete byte sequence, applies the
+caller-supplied mode, syncs and closes the file, runs an optional pre-rename
+check, replaces the destination with a rename, then syncs the containing
+directory.
+
+The pre-rename check is the last point at which a caller can cancel without
+installing a new destination. A failure before the rename leaves the old
+destination unchanged and removes the temporary file; any cleanup failure is
+returned alongside the primary failure. A failure after the rename may leave
+the new file visible, but it is not reported as crash-durable.
+
+Replacement follows the operating system's same-filesystem rename semantics.
+If a platform cannot replace an existing destination, the operation returns an
+error and never removes the old file as an emulation step.
+
+## Directory-Sync Support
+
+Linux and macOS attempt to sync the destination directory. Windows opens the
+directory with backup semantics and flushes its buffers. If either operation
+is unavailable for the platform, directory handle, or filesystem,
+`ErrDirectorySyncUnsupported` is returned. Narratio does not treat that result
+as successful crash-durable replacement.
+
+`WriteFileAtomic`, copy helpers, and downloaded temporary-file installation
+retain their compatibility behavior of creating the destination parent with
+the repository's workspace permissions before using this contract.

docs/internal/manifest.md

@@ -0,0 +1,157 @@
+# Internal: Manifest
+
+## Purpose
+
+Explain the session-progress and invocation-audit models implemented by
+`internal/manifest`. Physical manifest placement belongs in
+[Operations](../operations.md#local-state-layout).
+
+## Session Manifest
+
+`manifest.Manifest` records:
+
+- identity (`session_id`, `campaign`, `run_id`)
+- local path metadata (`local_workdir`, `local_spool_dir`)
+- remote identity metadata (`s3_bucket`, `s3_session_prefix`, `s3_run_prefix`)
+- `inputs` records
+- durable `artifacts` records
+- per-stage `stages` map
+- an optional `post_publish_cleanup` obligation, which binds a committed run,
+  remote commit identity, and each exact root-confined local target to its
+  completion evidence
+
+Session, campaign, and run identities in local and downloaded manifests must be
+portable opaque segments. Unsafe legacy identities are rejected with migration
+guidance rather than being normalized into a different workspace or remote
+namespace.
+
+The model admits these stage states:
+
+- `pending`
+- `running`
+- `succeeded`
+- `failed`
+- `skipped`
+- `stale`
+- `interrupted`
+
+## Run Manifest
+
+`manifest.RunManifest` is created for each invocation and records:
+
+- invocation identity and `force` flag
+- requested stages
+- per-stage action (`run` or `skip`)
+- per-stage status
+- overall run status (`running`, `succeeded`, `failed`)
+
+## Remote Commit Manifest
+
+`artifacts.RemoteCommitManifest` is a separate, versioned remote snapshot
+contract. It is not a serialized session manifest and contains no local
+post-publication assertion such as `current_pointer_written`. A remote commit
+identifies one campaign, session, and run and declares its immutable artifact
+set. Each artifact has a typed source, immutable destination key, SHA-256
+checksum, size, and storage generation.
+
+`current/commit-pointer.json` is the sole mutable selector for the new
+contract. It identifies exactly one run-scoped `runs/{run_id}/commit.json` and
+binds that object by checksum, size, and generation. Readers strictly reject
+unknown fields, version mismatches, pointer/commit identity mismatches, and
+objects that do not match their declaration.
+
+The reader retains a temporary, clearly isolated compatibility path for a
+coherent legacy `current/manifest.json` plus `current/run_id.txt` pair. That
+path is removable after migration and is never used to write new state.
+
+## Persistence Semantics
+
+`manifest.LocalStore`:
+
+- validates loaded documents;
+- normalizes missing maps/stage records;
+- writes through a sibling temporary file, syncing the completed file and
+  destination directory after atomic replacement;
+- updates `updated_at` on save.
+
+If the operating system or filesystem cannot sync a directory, save returns an
+explicit error instead of claiming crash-durable replacement. A returned error
+after the rename can therefore leave the new manifest visible but not confirmed
+durable; callers must reload it before retrying.
+
+## Execution Semantics
+
+The application runner marks an executing stage running and then succeeded or
+failed in both manifests, persisting each transition. On success it records
+outputs, logs, generated configuration references, and metadata. Artifact
+records may include optional contract and external provenance objects; old
+manifests remain compatible when those fields are absent. A successful forced
+rerun marks only succeeded downstream session-stage records stale.
+
+Starting an execution clears the current session-stage record's prior outputs,
+logs, generated configuration references, and metadata. Failed and skipped
+transitions enforce the same clearing rule directly, while success repopulates
+only fields returned by the new result. Marking a record stale does not clear
+those details because resume validation and diagnosis may still require them
+before execution begins. Invocation run manifests remain immutable audit
+records of their own outcomes.
+
+A stage may explicitly return a skipped disposition and stable reason. The
+runner persists that outcome in both manifests, clears older outputs for the
+session-stage record along with older logs, generated configuration references,
+and metadata, then applies any bounded details from the current skip and
+continues. This self-skip is distinct from deciding not to execute an
+already-succeeded stage and is reconsidered on later runs. Skipped results
+cannot contain outputs.
+
+When an already-succeeded stage is skipped, the invocation run manifest records
+the `skip` action and reason. The session manifest deliberately retains its
+existing succeeded record because it remains the cross-invocation progress
+authority. Stages with a resume validator, currently extraction, may reject an
+otherwise eligible skip when the recorded durable result is obsolete; the
+runner marks it stale and executes it.
+
+Session manifest is the authoritative stage-progress ledger across invocations.
+Run manifest is invocation-scoped audit state.
+
+After a publish commits remotely, any configured local cleanup is first recorded
+as a session-manifest obligation before deletion begins. Each target becomes
+complete only after its confined deletion (or safe absence check) and a
+successful manifest save. An incomplete obligation is retried on later
+invocations independently of their selected stages and retains the committed
+run and remote identity that authorized it.
+
+Each invocation derives campaign, session, run, local-path, and remote-prefix
+metadata from the validated resolved configuration as one projection. A persisted
+session manifest must agree on campaign and session identity before execution;
+the current projection is refreshed for every invocation while stage progress,
+inputs, and durable artifacts remain session history.
+
+For handled failures after an invocation record is created, the runner records
+the failure on the session ledger and persists it before persisting the failed
+run audit record. This preserves the resume authority while making a partial
+persistence disagreement visible. Abrupt process death remains an accepted case
+where a durable running record can require operator interpretation.
+
+## Invariants
+
+- stage resume/skip decisions are session-manifest driven.
+- running, failed, and self-skipped stages do not retain result payloads from
+  an earlier success.
+- stale stages retain prior details until replacement execution starts.
+- force reruns stale downstream succeeded stages.
+- run manifest does not replace session manifest as progress authority.
+- remote commitment is established by a verified current pointer and remote
+  commit relationship, never by a mutable session-manifest boolean.
+
+## Implementation And Tests
+
+- Models and transitions: `internal/manifest/manifest.go`,
+  `internal/manifest/run_manifest.go`
+- Remote commit model and readers: `internal/artifacts/remote_commit.go`,
+  `internal/artifacts/current_state_commit.go`,
+  `internal/artifacts/current_state_legacy.go`
+- Persistence and validation: `internal/manifest/store.go`
+- Package tests: `internal/manifest/*_test.go`
+- Assembled execution behavior: `internal/app/runner_test.go`,
+  `internal/app/run_stage_test.go`

docs/internal/overview.md

@@ -0,0 +1,98 @@
+# Internal Overview
+
+This document is the implemented component map for Narratio. Normative system
+boundaries and dependency direction belong in
+[Architecture](../policy/architecture.md). User and operator contracts belong
+in the [CLI](../cli.md), [Configuration](../config.md),
+[Operations](../operations.md), and [Troubleshooting](../troubleshooting.md).
+Externally observable tool and format contracts belong under
+[Integrations](../integrations/).
+
+## Execution Path
+
+```text
+cmd/narratio -> internal/app -> configuration and production composition
+             -> internal/stage -> adapters and external systems
+             -> manifests and artifact resolution -> durable local/remote output
+```
+
+The executable delegates process behavior to the application boundary. The
+application resolves configuration, composes concrete collaborators, acquires
+session safety controls, and runs commands. Pipeline commands execute the
+canonical stage sequence through adapter interfaces, while manifests record
+progress and artifact services resolve durable inputs and outputs.
+
+## Components
+
+| Area | Implemented owners | Responsibility |
+| --- | --- | --- |
+| Executable | `cmd/narratio` | Process entry, standard stream wiring, argument handoff, and exit status. |
+| Application orchestration | `internal/app` | Command dispatch, configuration selection, secret-file environment loading, production composition, session locking, planning, execution, restore, cleanup gates, and user-facing reporting. |
+| Configuration | `internal/config` | Strict YAML loading, discovery, defaults, normalization, session templating, and validation. |
+| Pipeline stages | `internal/stage` | Canonical stage registry, shared stage contract, execution dependencies, and implemented stage behavior. |
+| External boundaries | `internal/adapters`, `internal/audio` | WhisperX HTTP, downstream subprocesses, notification, object storage, and S3 audio materialization behind Narratio contracts. |
+| Manifests | `internal/manifest` | Durable session progress, invocation audit state, stage transitions, validation, and atomic persistence. |
+| Artifacts and paths | `internal/artifacts`, `internal/pathsafe` | Artifact identities and resolution, local and remote path/key models, current-state discovery, and confined relative destinations. |
+| Previous-session cache | `internal/previouscache` | Deterministic planning and materialization requirements for configured previous-session inputs. |
+| Artifact policy | `internal/artifactpolicy` | Source and destination policy, configured artifact identity validation, and publish destination safety. |
+| Shared models and file operations | `internal/artifactmodel`, `internal/contracts`, [`internal/fileops`](fileops.md) | Transcript and artifact data contracts plus durable single-file replacement helpers; unsupported directory syncing is reported explicitly. |
+| Logging | `internal/logging` | Application logger construction and shared structured logging behavior. |
+
+The application boundary composes concrete implementations. Stages depend on
+Narratio-level contracts; external transport and SDK details remain in
+adapters. The normative rules for these relationships remain in
+[Architecture](../policy/architecture.md).
+
+## Pipeline Stage Set
+
+The implemented canonical order is:
+
+1. [`prepare`](stage-prepare.md)
+2. [`transcribe`](stage-transcribe.md)
+3. [`merge`](stage-merge.md)
+4. [`polish`](stage-polish.md)
+5. [`normalize`](stage-normalize.md)
+6. [`trim`](stage-trim.md)
+7. [`extract`](stage-extract.md)
+8. [`render`](stage-render.md)
+9. [`analyze`](stage-analyze.md)
+10. [`publish`](stage-publish.md)
+11. `notify` (no-op)
+
+`notify` currently has no persisted pipeline outputs and uses the explicit
+`noop` notification mode. The focused stage documents own implementation
+mechanics. The
+[CLI](../cli.md) and [Operations](../operations.md) own user-visible invocation
+and execution semantics.
+
+## Focused Documentation
+
+- [Adapter Internals](adapters.md): external adapter boundaries, composition,
+  failure behavior, and test surfaces.
+- [Artifact Internals](artifacts.md): source identities, runtime catalog,
+  resolution, previous requirements, and current-state helpers.
+- [Manifest Internals](manifest.md): session and run records, persistence, and
+  execution transitions.
+- [Storage Internals](storage.md): object-store interface and S3 behavior.
+- [Workspace Internals](workspace.md): local layout, locking, and cleanup
+  guardrails.
+- [Restore Internals](command-restore.md): discovery, planning, execution, and
+  reporting.
+- [`prepare`](stage-prepare.md)
+- [`transcribe`](stage-transcribe.md)
+- [`merge`](stage-merge.md)
+- [`polish`](stage-polish.md)
+- [`normalize`](stage-normalize.md)
+- [`trim`](stage-trim.md)
+- [`extract`](stage-extract.md)
+- [`render`](stage-render.md)
+- [`analyze`](stage-analyze.md)
+- [`publish`](stage-publish.md)
+
+Use this map to find an owner, then read its focused documentation and tests
+before changing behavior.
+
+The stage registry is implemented in `internal/stage/placeholders.go` and its
+ordering is protected by `internal/app/planner_test.go`. Cross-invocation skip,
+force, failure, and invalidation behavior is exercised in
+`internal/app/runner_test.go` and `internal/app/run_stage_test.go`.

docs/internal/stage-analyze.md

@@ -0,0 +1,75 @@
+# Stage: analyze
+
+## Purpose
+
+Execute selected configured Scriptorium artifacts in dependency order and materialize outputs.
+
+## Inputs
+
+- configured artifacts from `pipeline.scriptorium.artifacts`
+- optional selected artifact keys supplied through the stage environment
+- built-in, configured, extraction, and previous-session source references in
+  artifact inputs
+
+Supported source families:
+- built-ins: `narratio.transcript.*`, `narratio.bounds.session`
+- prepared stable inputs: `narratio.input.players`, `narratio.input.party`,
+  `narratio.input.glossary`, `narratio.input.spell_catalog`
+- configured artifacts: `narratio.artifact.<key>`
+- extraction lanes: `narratio.extraction.<key>`
+- previous-session cache: `narratio.previous_session.artifact.<key>`
+
+## Outputs
+
+- one materialized output per executed configured artifact (`output_path`)
+- stage metadata describing selected/generated/reused artifacts
+
+## Key Behavior
+
+- when Scriptorium is absent or no configured artifact is executable, completes
+  successfully with no outputs and records explanatory metadata. This is not an
+  explicit self-skip: both manifests record success, satisfy publish's
+  prerequisite, and an ordinary later run reuses the result until forced.
+- builds a runtime artifact catalog containing built-ins, configured artifacts,
+  and configured extraction lanes. Extraction availability is hydrated only
+  from compatible successful extraction evidence.
+- uses enabled configured artifacts by default. An explicit `--artifacts`
+  selection is a one-invocation override: it makes exactly the named configured
+  artifacts executable even when disabled, and does not automatically include
+  dependencies. A selected artifact's dependencies must instead already be
+  available to the catalog.
+- marks non-executable configured artifacts as reusable when output files already exist.
+- validates selected artifact dependency order (cycle-safe topo ordering).
+- resolves required/optional inputs per artifact source definition.
+- omits an unavailable optional input; an unavailable required input fails.
+- resolves prepared stable input sources through the shared manifest-authoritative
+  identity resolver; it does not accept incidental files or fall back to
+  campaign/session source paths.
+- resolves previous-session sources from local `previous/` cache only.
+- runs optional render-debug, then artifact execution.
+- validates non-empty output files and materializes canonical outputs.
+
+## Failure Semantics
+
+- required missing configured/previous-session inputs fail.
+- missing required prepared stable input source includes prepare rerun guidance.
+- missing required previous-session source includes prepare rerun guidance.
+- missing required `narratio.transcript.final_markdown` or
+  `narratio.transcript.final_trimmed_markdown` inputs includes render rerun
+  guidance.
+- dependency cycles or unavailable required dependencies fail.
+- adapter validation failures fail stage.
+
+## Invariants
+
+- `analyze` performs no remote storage calls for previous-session source resolution.
+- output provenance and metadata are deterministic per execution.
+
+## Related Contracts And Tests
+
+- [Configuration](../config.md#scriptorium-artifact-entries) owns artifact
+  fields and source-selection rules.
+- [CLI](../cli.md) owns user-visible artifact selection.
+- [Scriptorium](../integrations/scriptorium.md) owns the subprocess contract.
+- Implementation and tests: `internal/stage/analyze.go`,
+  `internal/stage/analyze_test.go`

docs/internal/stage-extract.md

@@ -0,0 +1,109 @@
+# Internal: Extract Stage
+
+## Responsibility
+
+`extract` runs after `trim` and before `render`. It converts the canonical
+`narratio.transcript.final_trimmed` JSON into configured Notarius lane artifacts.
+An omitted or disabled Notarius section makes the stage explicitly self-skip
+with reason `notarius_disabled`, no outputs, and no Notarius runner.
+
+The external protocol is documented in the
+[Notarius integration contract](../integrations/notarius.md). Configuration
+fields belong in [Configuration](../config.md), and physical paths and force
+procedures belong in [Operations](../operations.md).
+
+## Lifecycle
+
+`internal/stage/extract.go`:
+
+1. resolves the final trimmed transcript from the shared artifact catalog;
+2. resolves every configured prepared reference through the shared
+   manifest-authoritative identity resolver before creating run-local output;
+3. streams each verified reference into an invocation-local snapshot and
+   rejects any source change observed while copying;
+4. fingerprints the Notarius invocation contract, including sorted reference
+   identities;
+5. creates a run-local staging directory and invokes the injected
+   `notarius.Runner`;
+6. revalidates the reference snapshots, then validates the v2 successful
+   receipt, confined index, management documents, configured required lane
+   descriptors, validation summaries, and regular payload files;
+7. atomically promotes the complete bundle to its immutable durable location;
+8. records one non-selectable `notarius_index` output and one selectable
+   `notarius_lane` output per configured lane; and
+9. registers each lane as `narratio.extraction.<output_key>` for downstream
+   Scriptorium and publish resolution.
+
+Lane records retain checksum, contract, producer run ID, and Notarius system,
+run, pipeline, and lane provenance. Stage metadata retains the durable bundle
+root, receipt, diagnostic paths, rejection/warning summaries, producing
+Narratio run ID, the resolved trimmed-input identity, and invocation
+fingerprint. The input identity binds the exact transcript bytes, canonical
+source ID, producer stage/output/run identity, and resolution provenance.
+Reference metadata contains only selector, source ID, canonical session-relative
+path, checksum, and size; adapter requests receive selector and absolute
+invocation-local snapshot path, never payload contents. Snapshot bytes must
+match the prepared identity both before and after Notarius runs, so a concurrent
+prepared-file replacement cannot make recorded provenance describe different
+bytes from those supplied to Notarius.
+Validation completes before
+promotion, so a rejected result cannot expose a partial durable bundle.
+
+Any executed extraction outcome that replaces a different effective outcome
+marks succeeded downstream stages stale. Repeating the same disabled self-skip
+with no outputs is stable and does not repeatedly invalidate downstream stages.
+
+## Resume Validation
+
+`internal/stage/extract_resume.go` permits a skip only when the existing stage
+record succeeded and still matches the current invocation fingerprint. The
+fingerprint covers the resolved executable and config paths, pipeline ID,
+timeout, working directory, sorted configured output contracts, the current
+direct trimmed-transcript identity, and sorted prepared-reference identities.
+The same reference helper and transcript identity are resolved again for
+artifact evidence, so changing the current transcript bytes or producer
+identity makes the prior extraction obsolete.
+
+A valid prepared-reference change makes extraction non-resumable. Missing,
+unsafe, or checksum-inconsistent prepared evidence is a hard validation error
+with prepare-force guidance because an immediate extract rerun cannot succeed.
+
+The validator then checks the producing run identity, canonical immutable
+bundle root, path confinement and absence of symlink components, receipt
+identity, exactly one canonical index, the exact configured source set,
+contracts and provenance, regular-file status, and stored checksums. Missing or
+obsolete results are non-resumable and run again; unsafe filesystem conditions
+return an error rather than silently accepting or replacing data.
+
+The fingerprint cannot observe files imported by Notarius configuration,
+profile contents, prompt/module definitions, or other transitive inputs.
+Operators must force extraction after changing any such input.
+
+## Failure Behavior
+
+Adapter startup, timeout, nonzero exit, receipt decoding, path confinement,
+index compatibility, inconsistent warning or diagnostic envelopes,
+required-lane rejection or incomplete validation, payload inspection,
+checksum, or promotion errors fail the stage through ordinary manifest
+transition handling.
+Stdout receipt and stderr diagnostics remain separate. Downstream stages are
+not given selectable extraction sources unless the complete configured result
+has passed validation and promotion.
+
+When a replacement attempt begins, the current session-stage record no longer
+advertises payload from the previous success. A failed replacement therefore
+has no current outputs, logs, generated configuration references, or metadata,
+while the earlier invocation manifest and immutable promoted bundle remain
+available for audit and recovery.
+
+## Implementation And Focused Tests
+
+- Stage execution, selection, and resume validation: `internal/stage/extract.go`,
+  `internal/stage/extract_resume.go`,
+  `internal/stage/extract_test.go`
+- Subprocess boundary: `internal/adapters/notarius/subprocess.go`,
+  `internal/adapters/notarius/subprocess_test.go`
+- Catalog hydration: `internal/artifacts/extraction_catalog.go`,
+  `internal/artifacts/extraction_catalog_test.go`
+- Composition and downstream behavior: `internal/app/runner_test.go`,
+  `internal/stage/analyze_test.go`, `internal/stage/publish_test.go`

docs/internal/stage-merge.md

@@ -0,0 +1,37 @@
+# Stage: merge
+
+## Purpose
+
+Normalize raw transcript inputs and merge into base transcript via Seriatim.
+
+## Inputs
+
+- `transcripts/raw/*.json`
+- `inputs/speakers.yml`
+- `inputs/autocorrect.yml`
+
+## Outputs
+
+- `transcripts/base.json`
+- optional `artifacts/seriatim.report.json`
+
+## Key Behavior
+
+- discovers and validates raw transcript inputs.
+- normalizes each raw transcript (`seriatim.Normalize`) into run-local scratch output.
+- merges normalized inputs (`seriatim.Run`) into base transcript.
+- validates merged transcript and optional report JSON.
+- materializes canonical outputs and records stage logs/generated configs.
+
+## Invariants
+
+- merge always consumes normalized forms of raw inputs.
+- base transcript must validate before stage success.
+- report output is config-gated.
+
+## Related Contracts And Tests
+
+- [Seriatim](../integrations/seriatim.md) owns subprocess and output semantics.
+- [Configuration](../config.md#pipeline) owns operator-selected Seriatim values.
+- Implementation and tests: `internal/stage/merge.go`,
+  `internal/stage/merge_test.go`

docs/internal/stage-normalize.md

@@ -0,0 +1,34 @@
+# Stage: normalize
+
+## Purpose
+
+Normalize polished transcript into final transcript using Seriatim.
+
+## Inputs
+
+- `transcripts/polished.json`
+
+## Outputs
+
+- `transcripts/final.json` (or configured normalize output path)
+- optional `artifacts/seriatim.normalize.report.json`
+
+## Key Behavior
+
+- resolves polished transcript from manifest outputs/canonical fallback.
+- applies `pipeline.normalize` config or default normalize config.
+- runs Seriatim normalize with configured timeout/binary.
+- validates normalized transcript and optional report.
+- materializes canonical outputs and records logs/generated configs.
+
+## Invariants
+
+- final transcript must validate as processed transcript JSON (`segments` array).
+- normalize defaults are applied when `pipeline.normalize` is unset.
+
+## Related Contracts And Tests
+
+- [Seriatim](../integrations/seriatim.md) owns subprocess and output semantics.
+- [Configuration](../config.md#pipeline) owns normalize fields and defaults.
+- Implementation and tests: `internal/stage/normalize.go`,
+  `internal/stage/normalize_test.go`

docs/internal/stage-polish.md

@@ -0,0 +1,37 @@
+# Stage: polish
+
+## Purpose
+
+Run Audita polishing on base transcript and produce polished transcript.
+
+## Inputs
+
+- `transcripts/base.json`
+- `inputs/glossary.yml`
+
+## Outputs
+
+- `transcripts/polished.json`
+- optional `artifacts/audita.report.json`
+
+## Key Behavior
+
+- resolves base transcript from merge outputs/canonical fallback.
+- invokes an Audita runner configured with static model/runtime options; the
+  invocation supplies paths and modules.
+- validates processed transcript structure (`segments` array required).
+- validates optional report JSON.
+- materializes canonical outputs; records logs/generated config and adapter metadata.
+
+## Invariants
+
+- polished transcript schema validation is mandatory.
+- report output is config-gated.
+
+## Related Contracts And Tests
+
+- [Audita](../integrations/audita.md) owns subprocess, validation, and failure
+  semantics.
+- [Configuration](../config.md#pipeline) owns operator-selected Audita values.
+- Implementation and tests: `internal/stage/polish.go`,
+  `internal/stage/polish_test.go`

docs/internal/stage-prepare.md

@@ -0,0 +1,70 @@
+# Stage: prepare
+
+## Purpose
+
+Materialize canonical current-session inputs before processing stages.
+
+## Inputs
+
+- resolved campaign, session, and pipeline configuration
+- stable input files (`speakers`, `autocorrect`, `glossary`, `players`, `party`)
+- optional spell-catalog overlay
+- one resolved local or S3 audio source
+- enabled configured artifact input requirements for previous-session sources
+
+## Outputs
+
+- `inputs/campaign.yml`
+- `inputs/session.yml`
+- `inputs/pipeline.resolved.yml`
+- `inputs/speakers.yml`
+- `inputs/autocorrect.yml`
+- `inputs/glossary.yml`
+- `inputs/players.yml`
+- `inputs/party.yml`
+- optional `inputs/spell_catalog.json`
+- `audio/*.flac`
+- optional `previous/manifest.json`
+- optional `previous/artifacts/**`
+- deterministic `manifest.inputs` entries (checksums + provenance)
+
+## Key Behavior
+
+- validates required config/store state.
+- enforces local audio vs S3 audio mutual exclusivity.
+- rejects duplicate explicit local audio sources after resolution.
+- gives distinct local source paths with the same basename deterministic unique
+  prepared filenames so neither source is overwritten.
+- materializes S3 audio through spool/cache-aware logic.
+- materializes a configured spell catalog with checksum and provenance, or
+  safely removes an obsolete canonical spell catalog and its manifest record
+  when the effective input is omitted.
+- scans enabled configured artifact inputs for `narratio.previous_session.artifact.*` requirements.
+- clears managed `previous/` state on every invocation, then, when requirements exist:
+  - resolves the pointer-selected previous source through the shared resolver;
+  - downloads previous manifest/artifacts;
+  - records previous inputs in `manifest.inputs`.
+
+Required previous-session inputs fail when unavailable; optional missing inputs
+are typed skipped results. Committed sources use their exact source-to-destination
+mapping, while the isolated legacy reader rejects ambiguous fallback matches.
+
+## Invariants
+
+- only `prepare` hydrates canonical `previous/` cache state.
+- managed previous artifacts are stored under `previous/artifacts/**` without
+  duplicate `artifacts/artifacts/` nesting.
+- managed `previous/` state represents only the current requirement set.
+- `manifest.inputs` ordering is deterministic (`kind`, `path`).
+
+## Related Contracts And Tests
+
+- [Configuration](../config.md) owns audio selection, stable input fields, and
+  previous-session settings.
+- [Operations](../operations.md) owns physical input, audio, spool, cache, and
+  previous-state layout.
+- [Storage Internals](storage.md) and [Artifact Internals](artifacts.md) explain
+  the internal collaborators.
+- Implementation and tests: `internal/stage/prepare.go`,
+  `internal/stage/prepare_test.go`, `internal/audio/s3_audio_test.go`,
+  `internal/previouscache/*_test.go`

docs/internal/stage-publish.md

@@ -0,0 +1,99 @@
+# Stage: publish
+
+## Purpose
+
+Upload run/session outputs to object storage and atomically advance remote current state.
+
+## Inputs
+
+- successful preceding stages from the [canonical stage set](overview.md#pipeline-stage-set)
+- invocation-scoped run files
+- resolved publish output rules
+- effective publish locks (static + remote merged lock set), revalidated at the
+  remote commit boundary
+- durable previous-session cache files when present
+
+## Outputs
+
+- uploaded invocation record and selected publish outputs;
+- uploaded durable previous-session cache files when present;
+- immutable run-scoped commit manifest; and
+- current commit pointer, written last.
+
+Exact remote placement and the operator workflow belong in
+[Operations](../operations.md#publish-workflow).
+
+## Key Behavior
+
+- when publishing or run upload is disabled, completes successfully with no
+  outputs and records explanatory metadata. This is not an explicit self-skip:
+  both manifests record success, and an ordinary later run reuses that result
+  until publish is forced.
+- validates prerequisite stage success and object-store availability.
+- derives a deterministic run-archive allowlist from the validated run
+  `manifest.json`: declared run-local outputs, logs, generated configs, and the
+  manifest itself. Unlisted workspace files are not archive candidates.
+- opens each archive candidate beneath its archive root without following
+  symlinked ancestors or leaf entries, verifies that it is a regular file and
+  checks a declared checksum when present, then streams the opened descriptor.
+- derives the durable previous-cache archive from its validated manifest using
+  the same confinement and regular-file checks.
+- resolves publish output sources through runtime artifact catalog and manifest-aware resolution.
+- publishes extraction lanes only through explicit configured output rules;
+  neither run-local nor durable Notarius bundles are scanned or uploaded wholesale.
+- selected artifact filter applies to configured artifact sources only.
+- locked outputs are skipped intentionally (including required ones).
+- optional missing outputs are skipped; required missing unlocked outputs fail.
+- creates one complete immutable source-to-destination mapping before upload;
+- uploads and verifies every declared immutable object and the commit manifest;
+- updates `current/commit-pointer.json` exactly once, last; and
+- does not write the legacy `current/manifest.json` or `current/run_id.txt` pair.
+- rechecks remote lock state immediately before the pointer update. A newly
+  committed lock aborts selection, leaving any uploaded immutable attempt
+  unselected.
+- reads the mutable remote lock document through a direct limit-plus-one read
+  capped by `MaxRemoteLockStoreBytes` (1 MiB), retaining the generation returned
+  with the opened body for conditional updates. Oversized lock documents fail
+  before YAML decoding; published artifact payloads do not use this limit.
+
+## Metadata Signals
+
+Includes counts/lists for:
+- run uploads
+- published output uploads
+- previous uploads
+- skipped optional outputs
+- skipped unselected outputs
+- locked outputs
+- remote commit and current-pointer key paths
+- the run identifier selected by the commit
+
+## Invariants
+
+- `current/commit-pointer.json` is the remote commit marker and is written last.
+- run files, selected outputs, previous-cache files, and the committed session
+  manifest are all declared by an immutable commit under the run prefix.
+- run and previous uploads contain only manifest-declared regular files opened
+  from verified descriptors; symlinks, special files, replacement races, and
+  undeclared entries are rejected or ignored before uploads begin.
+- run-local diagnostics, including Notarius receipt and stderr files, are
+  archived only when recorded by the run manifest.
+- publish locks are not overridden by `--force`; remote locks are revalidated
+  immediately before current-state selection.
+- post-commit local cleanup is authorized by the committed publish metadata and
+  is durably recorded by the application lifecycle before any local deletion.
+
+The commit boundary and cleanup gate are normative architecture invariants; see
+[Architecture](../policy/architecture.md#publish-commit-boundary).
+
+## Related Contracts And Tests
+
+- [Configuration](../config.md#publish-configuration-summary) owns output and
+  static-lock fields.
+- [Operations](../operations.md#publish-locks) owns remote lock lifecycle and
+  physical remote state.
+- [Artifact Internals](artifacts.md) explains source resolution and current-state
+  helpers.
+- Implementation and tests: `internal/stage/publish.go`,
+  `internal/stage/publish_test.go`, `internal/app/operator_helpers_test.go`,
+  `internal/app/post_publish_cleanup_test.go`

docs/internal/stage-render.md

@@ -0,0 +1,44 @@
+# Stage: render
+
+## Purpose
+
+Render Markdown transcript artifacts from normalized JSON transcripts via Seriatim.
+
+## Inputs
+
+- `narratio.transcript.final` (`transcripts/final.json`)
+- `narratio.transcript.final_trimmed` (`transcripts/final.trimmed.json`)
+
+## Outputs
+
+- `narratio.transcript.final_markdown` -> `transcripts/final.md`
+- `narratio.transcript.final_trimmed_markdown` -> `transcripts/final.trimmed.md`
+
+## Key Behavior
+
+- uses `pipeline.render` settings (enabled/format/title/booleans).
+- resolves inputs manifest-first, then canonical fallback.
+- writes run-local outputs first, then materializes canonical session outputs.
+- records input provenance, output paths, adapter metadata, logs, and generated config refs.
+- when `pipeline.render.enabled=false`, completes successfully with no outputs
+  and records explanatory metadata. This is not an explicit self-skip: both
+  manifests record success, and enabling render later requires a forced run.
+
+## Failure Semantics
+
+- missing normalized input fails with normalize rerun guidance.
+- missing trimmed input fails with trim rerun guidance.
+- adapter/subprocess failure fails stage.
+- empty render output files fail validation.
+
+## Invariants
+
+- only `format: markdown` is supported.
+- render stage owns production of built-in Markdown transcript sources.
+
+## Related Contracts And Tests
+
+- [Seriatim](../integrations/seriatim.md) owns render subprocess behavior.
+- [Configuration](../config.md#pipeline) owns render fields and defaults.
+- Implementation and tests: `internal/stage/render.go`,
+  `internal/stage/render_test.go`

docs/internal/stage-transcribe.md

@@ -0,0 +1,41 @@
+# Stage: transcribe
+
+## Purpose
+
+Generate raw per-speaker transcripts from prepared audio using WhisperX.
+
+## Inputs
+
+- `audio/*.flac` from `prepare`
+
+## Outputs
+
+- `transcripts/raw/<speaker>.json`
+
+## Key Behavior
+
+- discovers prepared audio from manifest inputs or canonical audio directory.
+- derives the transcript identity from the prepared `.flac` filename.
+- dispatches WhisperX requests through a bounded worker pool.
+- validates each output as JSON.
+- writes run-local outputs then materializes canonical transcript outputs only
+  after every planned request succeeds.
+
+## Invariants
+
+- prepared audio identities must be unique; prepare disambiguates distinct
+  source paths that share a basename.
+- output path returned by adapter must match requested output path.
+- an empty adapter result path means the requested path; adapters cannot select
+  an alternate destination.
+- each successful output is validated before stage success, and cancellation or
+  incomplete dispatch cannot be reported as a successful result.
+
+## Related Contracts And Tests
+
+- [WhisperX](../integrations/whisperx.md) owns HTTP, retry, timeout, and
+  cancellation semantics.
+- [Configuration](../config.md#pipeline) owns concurrency and other
+  operator-selected values.
+- Implementation and tests: `internal/stage/transcribe.go`,
+  `internal/stage/transcribe_test.go`

docs/internal/stage-trim.md

@@ -0,0 +1,43 @@
+# Stage: trim
+
+## Purpose
+
+Produce a final-trimmed transcript. By default, the stage generates bounds and
+applies a bounds-driven trim.
+
+## Inputs
+
+- `transcripts/final.json`
+
+## Outputs
+
+- `transcripts/final.trimmed.json` (or configured trim output path)
+- when trim enabled: `artifacts/session_bounds.json`
+
+## Key Behavior
+
+When `trim.enabled=true`:
+- runs Scriptorium bounds artifact generation;
+- optionally runs render-debug output generation;
+- validates bounds payload against transcript;
+- derives keep selector;
+- either copies unchanged transcript or runs Seriatim trim;
+- validates trimmed transcript and materializes bounds output.
+
+When `trim.enabled=false`:
+- copies normalized transcript to trimmed output.
+
+## Invariants
+
+- normalized transcript is required input.
+- bounds output exists only in enabled trim path.
+- render-debug output is diagnostic and not a declared stage output.
+
+## Related Contracts And Tests
+
+- [Scriptorium](../integrations/scriptorium.md) owns bounds generation and
+  debug-render subprocess behavior.
+- [Seriatim](../integrations/seriatim.md) owns transcript trimming behavior.
+- [Configuration](../config.md#pipeline) owns trim fields and defaults.
+- Implementation and tests: `internal/stage/trim.go`,
+  `internal/stage/trim_test.go`

docs/internal/storage.md

@@ -0,0 +1,64 @@
+# Internal: Storage
+
+## Purpose
+
+Explain the object-store interface and S3 implementation used by Narratio.
+Remote key layout and lifecycle belong in [Operations](../operations.md), while
+operator-selected storage fields and credential mechanisms belong in
+[Configuration](../config.md).
+
+## Primary Contract
+
+`storage.ObjectStore` interface:
+
+- `List(ctx, prefix)`
+- `Read(ctx, key)` returns an object body and the generation observed with it
+- `Download(ctx, key, localPath)`
+- `Upload(ctx, localPath, key, opts)`
+- `UploadConditional(ctx, source, key, opts, condition)`
+- `Exists(ctx, key)`
+
+Key invariant:
+- callers pass full bucket-relative keys;
+- storage implementations do not infer campaign/session/run prefixes.
+
+`ReadObjectBounded` is the shared mechanism for small control objects. It opens
+one object version, returns the metadata observed with that body, rejects an
+oversized known size before transfer, and still performs a context-aware
+limit-plus-one read. It closes the body on every exit. Callers own the policy
+limit and add the control-object category to errors; this helper is not used for
+large artifact payloads.
+
+## Composition
+
+`NewObjectStoreFromConfig` constructs the S3-backed implementation from
+resolved configuration. The application loads configured filesystem secrets
+before calling it. The storage adapter consumes already-resolved values; it does
+not own discovery, defaults, or configuration validation.
+
+## S3 Backend Behavior
+
+- normalizes object keys.
+- `List` paginates and returns normalized `ObjectInfo`.
+- A truncated S3 listing must supply a new, non-empty continuation token;
+  otherwise listing fails with bucket and prefix context instead of looping.
+- `Download` writes local files with parent directory creation.
+- `Upload` streams local file and returns remote metadata.
+- `Read` binds a returned body to its S3 ETag. `UploadConditional` maps an ETag
+  match or absence precondition directly to the provider request and reports a
+  failed precondition without performing a local check-then-write replacement.
+- `Exists` maps not-found responses to `false`.
+
+## Invariants
+
+- storage layer is stateless regarding manifest/stage progression.
+- bounded reads never retain more than the caller's limit plus one byte and do
+  not replace owner-specific size policy.
+- publish ordering semantics are owned by stage/app code, not storage adapters.
+
+## Implementation And Tests
+
+- Contract and S3 adapter: `internal/adapters/storage`
+- Composition: `internal/app/object_store.go`
+- Tests: `internal/adapters/storage/*_test.go`,
+  `internal/app/object_store_test.go`

docs/internal/workspace.md

@@ -0,0 +1,94 @@
+# Internal: Workspace
+
+## Purpose
+
+Explain the helpers that construct local session and run paths, coordinate
+single-writer access, and confine cleanup. The authoritative physical layout and
+retention workflow belong in [Operations](../operations.md#local-state-layout).
+
+## Path Ownership
+
+`internal/artifacts` owns canonical session, run, spool, cache, and
+previous-cache path construction. `SessionPathsFor` provides the session-scoped
+path model, and layout creation goes through `EnsureLayoutFor`. Callers should
+consume those helpers instead of rebuilding relative paths.
+
+`internal/pathsafe` validates relative destinations. `internal/fileops` opens
+cleanup roots and their descendants through no-follow directory handles before
+removing them.
+
+`internal/fileops` owns the ordinary workspace mode contract. On POSIX,
+`WorkspaceDirectoryMode` is setgid `02775` and `WorkspaceFileMode` is `0664`.
+`EnsureWorkspaceDirectory` reapplies the directory mode after creation so a
+restrictive umask cannot remove group access, while retaining existing ownership
+and group. Credential paths are outside this contract; the platform-specific
+operational requirements are in [Operations](../operations.md#workspace-permissions).
+
+## Run-Local Stage Layout
+
+`internal/stage/run_local.go` maps stage outputs and diagnostics into an
+invocation-scoped layout. Successful outputs are validated and atomically
+materialized into canonical session paths before stage success. Managed
+previous-session cache paths remain session-durable and are never redirected
+into run-local output space.
+
+Extraction uses run-local receipt, stderr, and output-root helpers, then
+promotes the validated external bundle to the unique immutable Notarius bundle
+path supplied by `internal/artifacts`. `internal/fileops.PromoteDirectory`
+copies only regular files and directories to a same-filesystem temporary
+sibling. Source traversal uses confined directory handles and identity checks
+so replacing an inspected root, directory, or file is rejected rather than
+followed. The completed tree is atomically renamed without replacing an
+existing destination. Exact physical paths belong in
+[Operations](../operations.md#extraction-workflow).
+
+## Locking
+
+`artifacts.LocalStore` enforces the single-writer session lock via an
+operating-system lock held on `.lock` (`ErrLockConflict` on contention). The
+file retains owner metadata after release or process death; its existence is
+not evidence that a lock is active. Command and restore flows wait for this
+lock only while their context remains active, and report a release failure.
+
+## Cleanup Semantics
+
+Automatic post-publish cleanup:
+
+- is created only after a successful publish commit with complete publish
+  metadata, then is persisted before any deletion;
+- requires `uploaded=true`, a remote commit key, and a current commit-pointer
+  key in publish metadata;
+- consumes the resolved cleanup policy described in
+  [Configuration](../config.md);
+- refuses unsafe deletes (root delete, out-of-root delete, and symlinked
+  ancestors or entries);
+- retries any recorded incomplete target on later invocations even when no
+  publish work is selected. Missing targets are a successful, idempotent
+  cleanup result only after the completion evidence is saved.
+
+Manual cleanup uses the same root-confined deletion mechanism. Invocation
+syntax and exact deletion scope belong in [CLI](../cli.md#clean) and
+[Operations](../operations.md#cleanup).
+
+## Invariants
+
+- campaign-aware session root is mandatory.
+- manifest-driven stage state is durable across runs.
+- cleanup guardrails prevent destructive root/out-of-scope deletion.
+- ordinary workspace paths retain group-writable directory and file modes across
+  nested creation, replacement, and Notarius promotion.
+
+## Implementation And Tests
+
+- Path model and local store: `internal/artifacts/paths.go`,
+  `internal/artifacts/local.go`
+- Run-local materialization: `internal/stage/run_local.go`
+- Immutable bundle promotion: `internal/fileops/directory.go`
+- Workspace modes: `internal/fileops/modes.go`
+- Cleanup confinement: `internal/fileops/cleanup.go`,
+  `internal/app/cleanup_targets.go`, `internal/app/post_publish_cleanup.go`
+- Tests: `internal/artifacts/paths_model_test.go`,
+  `internal/artifacts/local_test.go`, `internal/stage/run_local_test.go`,
+  `internal/fileops/directory_test.go`, `internal/fileops/modes_posix_test.go`,
+  `internal/fileops/cleanup_test.go`, `internal/app/cleanup_targets_test.go`,
+  `internal/app/post_publish_cleanup_test.go`

docs/operations.md

@@ -0,0 +1,447 @@
+# Operations Guide
+
+Operator workflow for running, recovering, and publishing Narratio sessions.
+
+For command syntax, see [docs/cli.md](./cli.md). For field-level config, see [docs/config.md](./config.md).
+
+## Campaign and Session Selection
+
+Campaign selection priority:
+
+- `--campaign-file`
+- `--campaign`
+- `pipeline.campaigns.default_campaign_id`
+
+Session source priority:
+
+- `--session`
+- local default search paths
+- remote session object (S3) when local session file is not found and storage is configured
+
+## Session Initialization
+
+Use `session init` to generate a concrete session file for local or remote use.
+
+Local file:
+
+```bash
+narratio session init 2026-04-04 --output ./session.yml --date 2026-04-04 --title "Session 12"
+```
+
+Remote session object:
+
+```bash
+narratio session init 2026-04-04 --remote --force
+```
+
+If `campaign.yml` sets `session_template_file`, `session init` renders it. Template variables must resolve to concrete values.
+
+Campaigns must provide stable input files for speakers, autocorrect, glossary,
+players, and party, and may provide an optional spell-catalog overlay. Session
+files may override those paths for one session. The `prepare` stage
+materializes them under `inputs/`; configured consumers use the prepared files,
+never the original campaign or session source paths. Field definitions and
+source IDs are in [Configuration](./config.md#notarius-reference-bindings).
+
+## Standard Session Workflow
+
+1. Select pipeline/campaign/session config.
+2. Validate session readiness:
+
+```bash
+narratio session validate 2026-04-04
+```
+
+3. (Optional) inspect stage decisions:
+
+```bash
+narratio session plan 2026-04-04
+```
+
+4. Run the pipeline:
+
+```bash
+narratio run 2026-04-04
+```
+
+5. Check state:
+
+```bash
+narratio session status 2026-04-04
+```
+
+## Stage Execution and Continuation Behavior
+
+Canonical stage order:
+
+1. `prepare`
+2. `transcribe`
+3. `merge`
+4. `polish`
+5. `normalize`
+6. `trim`
+7. `extract`
+8. `render`
+9. `analyze`
+10. `publish`
+11. `notify`
+
+Execution rules:
+
+- succeeded stages are skipped unless `--force` is set;
+- `run` continues interrupted or partially completed sessions by running non-succeeded stages;
+- forcing an upstream stage marks succeeded downstream stages as `stale` before
+  the replacement runs; and
+- an executed failure, changed self-skip, or success that replaces a different
+  effective upstream outcome also marks succeeded downstream stages stale. A
+  repeated self-skip with the same reason and no outputs is stable and does not
+  perpetually rerun downstream work.
+
+An explicit self-skip is a durable `skipped` stage outcome that later runs
+reconsider. It differs from successful no-output execution: disabled `render`
+and `publish`, and absent or no-executable `analyze`, record `succeeded` with
+metadata and no outputs. Ordinary later runs reuse those successful results;
+force the affected stage after enabling or configuring it. Optional artifact
+inputs are omitted only from the consuming artifact invocation and do not make
+the stage self-skip.
+
+Single-stage execution:
+
+```bash
+narratio run-stage normalize 2026-04-04 --force
+```
+
+## Artifact Selection
+
+`--artifacts` can be used on `run`, `run-stage`, `analyze`, and `publish`.
+
+Selection behavior:
+
+- validates names against `pipeline.scriptorium.artifacts`;
+- filters analyze execution to selected configured artifacts;
+- filters publish rules for `narratio.artifact.<name>` sources only;
+- does not suppress built-in transcript, bounds, or explicitly configured
+  `narratio.extraction.<name>` publish sources; and
+- never partially selects Notarius lanes.
+
+## Extraction Workflow
+
+When Notarius is omitted or disabled, `extract` records an explicit skipped
+outcome with reason `notarius_disabled` and no outputs. A later invocation
+reconsiders the skipped stage, so enabling Notarius does not require force.
+
+When Notarius extraction is enabled, the stage consumes the final trimmed JSON
+and preserves the complete validated Notarius bundle at:
+
+- `artifacts/notarius/{narratio_run_id}/`
+
+The directory is immutable once promoted. Configured lanes become
+`narratio.extraction.<name>` sources for Scriptorium and explicit publish rules;
+the bundle and `index.json` are retained for audit and resume validation but
+are not selectable or published implicitly.
+
+Configured Notarius references resolve only from the current manifest-backed
+prepared inputs. Their canonical locations are `inputs/party.yml`,
+`inputs/players.yml`, `inputs/glossary.yml`, and, when configured,
+`inputs/spell_catalog.json`. Extraction supplies Notarius with verified copies
+under `runs/<run_id>/extract/references/` so a concurrent refresh of canonical
+prepared files cannot change the bytes consumed by an in-flight invocation.
+Inspect the effective stable-input inventory and
+prepared-file readiness with:
+
+```bash
+narratio session status 2026-04-04
+narratio session validate 2026-04-04
+```
+
+Reference metadata records selector, source ID, session-relative path,
+checksum, and byte size, but never payload contents. Changing a prepared
+reference changes extraction identity: ordinary continuation rejects the old
+result, reruns Notarius, and marks successful downstream stages stale. If the
+prepared file is missing or inconsistent with its manifest checksum, repair
+the source configuration and refresh prepared state first:
+
+```bash
+narratio run-stage prepare 2026-04-04 --force
+```
+
+Starting a replacement clears the previous extraction payload from the current
+session-stage record. If that replacement fails or self-skips, the current
+record does not fall back to the earlier outputs. The earlier run manifest and
+immutable bundle remain available for inspection, but downstream resolution
+requires a new current successful extraction record.
+
+Atomic Notarius bundle promotion is supported on Linux and macOS. On Windows
+and other operating systems, extraction fails before copying the bundle into a
+temporary promotion tree because Narratio has no verified atomic no-replace
+directory primitive there. This is an extraction limitation, not a broader
+platform-support guarantee for every Narratio workflow.
+
+## External Command Lifecycle
+
+When an external command is cancelled or times out, Narratio terminates its
+owned descendants as well as the command itself. Cancellation first requests
+termination where the platform supports it, then force terminates after a
+bounded wait. A command is not considered finished until its leader has been
+reaped, and descendants that keep standard output or error open cannot keep
+the invocation blocked. Other operating systems fail closed rather than launch
+a command without tree ownership.
+
+Subprocess stdout and stderr diagnostics are separately redacted and capped at
+8 MiB per invocation. Narratio does not retain configured credential values in
+these logs or their error tails; reaching a capture limit terminates the command
+tree and reports which stream exceeded the limit.
+
+Run-local diagnostics are:
+
+- `runs/{run_id}/extract/notarius.receipt.json`
+- `runs/{run_id}/extract/notarius.stderr.log`
+- `runs/{run_id}/extract/notarius-output/` before durable promotion
+
+The run-record upload is an allowlist derived from the validated run manifest,
+not a workspace scan. Each declared source is opened without following
+symlinked ancestors or the leaf, verified as a regular file, and streamed from
+that verified descriptor. Unlisted files and unsafe entries are never uploaded.
+The durable bundle is never scanned for implicit publication; only lanes named
+by explicit `pipeline.publish.outputs` rules are uploaded.
+
+To intentionally replace the current extraction result, run:
+
+```bash
+narratio run-stage extract 2026-04-04 --force
+```
+
+Narratio automatically reruns extraction when its recorded invocation contract,
+prepared Narratio reference identities, or durable output validation changes.
+It cannot fingerprint configuration files, profiles, prompts, modules, or
+other references loaded transitively by Notarius itself. Force extraction after
+changing any of those inputs, even when the top-level Narratio and Notarius
+config paths remain the same. A forced extract
+marks successful downstream stages stale. Ordinary extraction failures or
+outcome changes also stale affected downstream stages, while an identical
+repeated `notarius_disabled` self-skip does not repeatedly invalidate them.
+
+## Publish Workflow
+
+Run publish only:
+
+```bash
+narratio publish 2026-04-04
+```
+
+Equivalent:
+
+```bash
+narratio run-stage publish 2026-04-04 --force
+```
+
+Publish commit model:
+
+- uploads eligible run files under `{session_prefix}/runs/{run_id}/`, excluding
+  audio and the run-local Notarius staging bundle;
+- uploads configured published outputs and `previous/**` cache files into the
+  same immutable run scope, including only explicitly configured extraction
+  lanes;
+- writes `{session_prefix}/runs/{run_id}/commit.json` after all declared
+  immutable objects are uploaded and verified; and
+- writes `{session_prefix}/current/commit-pointer.json` once, last.
+
+`current/commit-pointer.json` is the remote current-state commit marker. It
+selects exactly one immutable commit, which declares the complete object set.
+
+## Remote Commit Migration
+
+The immutable remote commit contract uses
+`runs/{run_id}/commit.json` to declare a run's complete object set and a small
+`current/commit-pointer.json` to select it. The pointer binds the selected
+commit by version, checksum, size, and storage generation; committed artifacts
+are also checksum- and generation-bound. Readers accept this contract now and
+strictly reject mismatched or unknown data.
+
+Legacy reads are limited to a coherent `current/manifest.json` and
+`current/run_id.txt` pair; a torn pair is rejected. New publication does not
+write that pair and remote commit state does not carry local
+`current_pointer_written` metadata.
+
+## Publish Locks
+
+Lock sources:
+
+- static locks in `pipeline.publish.locks`
+- mutable remote locks in `{session_prefix}/locks.yml`
+
+Effective lock rules:
+
+- static and remote locks are merged;
+- static locks win on source collisions;
+- locked outputs are intentional skips;
+- lock add/remove commands mutate only remote lock state through generation-bound
+  conditional writes. A command retries a bounded number of concurrent
+  conflicts while its invocation context remains active, so it never replaces a
+  different lock-document generation; and
+- a publish re-reads remote locks immediately before it writes the current
+  commit pointer. A lock committed before that recheck prevents selecting the
+  new snapshot, even though its already-uploaded immutable objects may remain
+  available for a later retry.
+
+Examples:
+
+```bash
+narratio session locks 2026-04-04
+narratio session locks add 2026-04-04 narratio.artifact.session_recap --reason "manual edits" --force
+narratio session locks remove 2026-04-04 narratio.artifact.session_recap
+```
+
+## Restore Workflow
+
+Use restore when local durable session state is missing or stale and remote committed current state is authoritative.
+
+Dry run:
+
+```bash
+narratio session restore 2026-04-04 --dry-run
+```
+
+Apply:
+
+```bash
+narratio session restore 2026-04-04
+```
+
+`--dry-run` does not write durable session files. It still reads the selected
+remote current state and may read object identity/content needed to classify the
+plan, so it is not a network-free operation.
+
+Default restore scope:
+
+- the committed session manifest and the committed transcript/artifact objects
+  declared by the selected remote commit
+- `previous/**` when needed by configured previous-session artifact inputs
+
+Optional:
+
+- `--include-audio` to include `audio/**`
+- `--force` to overwrite eligible conflicting regular files; it never replaces
+  directories or other non-regular local targets
+
+Restore writes an execution report at `reports/restore-latest.json`.
+
+If restore fails after beginning installation, it leaves a durable
+`.restore-incomplete.json` marker in the session root. Pipeline runs will stop
+until you rerun the same restore command and it completes. Restore intentionally
+does not try to roll back files already installed; retrying the selected remote
+snapshot is the recovery procedure.
+
+## Local State Layout
+
+Session root:
+
+- `{workspace.root}/work/{campaign}/{session_id}`
+
+Durable session paths:
+
+- `manifest.json`
+- `inputs/**`
+- `audio/**`
+- `transcripts/**`
+- `artifacts/**`
+- `previous/**`
+- `reports/**`
+- `logs/**`
+- `config/**`
+- `runs/**`
+
+Validated Notarius bundles live below `artifacts/notarius/{run_id}/`; receipt,
+stderr, and pre-promotion output remain in the producing run's `extract`
+directory as described in [Extraction Workflow](#extraction-workflow).
+
+Run-local layout:
+
+- `runs/{run_id}/{stage}/outputs`
+- `runs/{run_id}/{stage}/logs`
+- `runs/{run_id}/{stage}/reports`
+- `runs/{run_id}/{stage}/config`
+- `runs/{run_id}/{stage}/scratch`
+
+Spool layout (runtime/transient):
+
+- `{spool.root}/{campaign}/{session_id}/{run_id}/...`
+- restore audio spool under `{spool.root}/{campaign}/{session_id}/restore/audio`
+
+Cache layout (durable S3 audio cache):
+
+- `{cache.root}/s3/{bucket}/...`
+
+Each cached audio file has an adjacent managed identity record. It binds the
+file to its remote object version and verified digest; deleting or altering the
+record simply causes Narratio to download and verify the object again.
+
+### Workspace Permissions
+
+Ordinary Narratio workspace content is intentionally shareable with the
+workspace group. On POSIX systems, Narratio-created workspace, spool, and cache
+directories converge on setgid `02775`; ordinary files, including manifests,
+transcripts, generated configuration, logs, reports, and Notarius artifacts,
+converge on `0664`. Narratio explicitly applies these modes so a restrictive
+caller umask does not remove group write or setgid. It does not change file or
+directory ownership: the configured workspace's existing group is inherited.
+
+Windows does not implement POSIX mode bits or setgid semantics. Configure the
+workspace, spool, and cache locations with an ACL that grants the collaborating
+group read/write access, and configure credential locations with an ACL limited
+to the intended credential owner. Do not use POSIX mode displays as evidence of
+Windows access control.
+
+API keys are credentials, not ordinary workspace data. Store them outside the
+shared workspace or in a separately restricted credential location; ordinary
+workspace group access must never be treated as authorization to read keys.
+On POSIX, provision a credential directory as `0700` and credential files as
+`0600`; Narratio rejects group- or other-readable configured credential paths.
+On Windows, restrict the directory and files with ACLs to the credential owner.
+
+External adapter results are individually bounded before Narratio validates or
+materializes them. These per-file limits do not reserve disk space: prevent hard
+disk exhaustion with filesystem, service, container, or volume quotas sized for
+the session workload.
+
+## Cleanup
+
+Session-scoped cleanup:
+
+```bash
+narratio clean 2026-04-04
+```
+
+Global cleanup:
+
+```bash
+narratio clean --all
+```
+
+Dry-run and cache variants:
+
+```bash
+narratio clean 2026-04-04 --dry-run --clear-cache
+narratio clean --all --dry-run --clear-cache
+```
+
+Rules:
+
+- `clean` deletes work/spool session state;
+- cache is preserved unless `--clear-cache` is set;
+- each deletion is confined beneath its configured workspace, spool, or cache
+  root and refuses symlinked paths;
+- automatic post-publish cleanup is gated by successful publish commit plus:
+  - `pipeline.spool.delete_audio_after_publish=true`
+  - `pipeline.workspace.cleanup_after_publish=true`
+  - Narratio first records the exact run-scoped cleanup obligation. If cleanup
+    reports incomplete, the remote committed snapshot remains current; rerun
+    Narratio to retry only the outstanding confined local cleanup.
+
+## Operational Caveats
+
+- Local and S3 audio modes are mutually exclusive.
+- Publish requires prerequisite stages through `render` and `analyze` to be succeeded.
+- Markdown publish defaults require render outputs (`transcripts/final.md` and `transcripts/final.trimmed.md`).
+- Restore requires configured object storage and committed remote current state.
+- Storage-backed commands load filesystem secrets before object-store initialization.

docs/policy/architecture.md

@@ -0,0 +1,272 @@
+# Architecture
+
+This document defines Narratio's intended high-level architecture and the
+invariants that changes must preserve. Implemented component details belong in
+the [Internal Overview](../internal/overview.md) and its linked documents.
+Significant architectural decision history belongs under `docs/adr/` when such
+records exist.
+
+## System Shape
+
+Narratio is a small Go application that turns D&D session audio into polished
+transcripts and generated session artifacts. It is an explicit, stage-driven
+orchestrator, not a general workflow engine.
+
+Narratio coordinates specialized external systems rather than reimplementing
+their domains:
+
+- WhisperX performs transcription;
+- Seriatim performs deterministic transcript processing and rendering;
+- Audita performs transcript correction and polishing;
+- Notarius extracts validated structured artifact bundles; and
+- Scriptorium executes prompts and produces configured artifacts.
+
+Narratio owns orchestration, configuration resolution, session and run state,
+artifact and path modeling, manifest persistence, stage sequencing, resume,
+restore, cleanup gates, and publish semantics. External contracts are defined
+in the [integration documentation](../integrations/).
+
+The pipeline has one canonical ordered stage set. Configuration may enable,
+disable, or parameterize supported behavior, but it must not turn that sequence
+into an arbitrary DAG or hide orchestration in generic workflow abstractions.
+The implemented stage inventory belongs in the
+[Internal Overview](../internal/overview.md).
+
+Narratio is contract-first without being abstraction-heavy. Interfaces and
+extension points should protect demonstrated boundaries. New abstraction is not
+itself an architectural goal.
+
+## Ownership And Dependency Direction
+
+The application boundary owns command dispatch, configuration selection,
+production composition, session locking, and top-level lifecycle. It may depend
+on concrete implementations to assemble a run.
+
+Stage orchestration expresses intent in Narratio-level data and interfaces.
+Stages may depend on configuration, manifest, artifact, path, and adapter
+contracts, but they must not depend on transport-specific request types,
+subprocess argument construction, cloud SDK types, or downstream tool internals.
+
+Adapters translate between Narratio contracts and external systems. They own
+HTTP, subprocess, notification, and object-storage mechanics, including command
+construction, transport behavior, provider response handling, and external
+error adaptation. External dependency types must remain inside the adapter that
+owns them unless that dependency is the adapter's explicit public contract.
+WhisperX HTTP behavior, Seriatim, Audita, Notarius, and Scriptorium command
+construction, notification transport, and object-storage SDK details remain
+behind these boundaries.
+
+State and path services must not infer stage policy. Storage implementations
+receive explicit bucket-relative keys and do not infer campaign, session, run,
+or root-prefix semantics. Manifest persistence records transitions but does not
+choose orchestration policy. Artifact resolution identifies and validates
+artifacts but does not execute producers.
+
+Dependencies should remain narrow and point toward Narratio-owned contracts.
+Prefer the Go standard library. Add an external dependency only when it provides
+a clear correctness, security, interoperability, or complexity benefit, and
+confine it to the boundary that needs it.
+
+## Stage Boundaries
+
+Each stage has one explicit responsibility and declares:
+
+- required input state;
+- produced output state;
+- configuration it consumes;
+- external adapters it uses;
+- manifest references and metadata it reads or writes;
+- skip, force, invalidation, and resume behavior; and
+- failure behavior.
+
+Stages write and validate run-local results before materializing canonical
+outputs where that distinction applies. A stage is complete only after its
+required outputs have been written, validated, and recorded in durable manifest
+state. Later stages depend on recorded success and artifact resolution, not
+merely on incidental files existing on disk.
+
+A failed or interrupted stage must not be presented as successful. Failure
+should preserve enough local state and diagnostics for inspection, recovery,
+and resume. Forcing an upstream stage invalidates succeeded downstream work
+according to the canonical stage order.
+
+A stage may explicitly self-skip with a stable reason and no outputs. That
+outcome is persisted, clears older outputs owned by the stage, and is
+reconsidered on a later invocation. A stage may also validate whether an
+otherwise successful recorded result is still resumable; an obsolete result
+is staled and rerun, while an unsafe condition that prevents a sound decision
+stops execution.
+
+Shared behavior should live behind a narrow service or helper with one clear
+owner. Stages must not reach across boundaries or reproduce adapter, manifest,
+artifact, or path policy ad hoc.
+
+## Manifest, Resume, And Restore
+
+The session manifest is the durable ledger for progress across invocations. It
+records session and run identity, stage state, input and output references,
+diagnostic references, checksums or provenance where useful, and non-secret
+adapter and publish metadata.
+
+Resume and skip decisions are manifest-driven. Filesystem state may be
+inspected and validated, but file presence alone does not replace recorded
+stage state. Invocation-scoped run records provide an audit of one execution;
+they do not replace the session manifest as progress authority.
+
+Restore treats committed remote current state as its authority. It must plan
+deterministically, confine remote-to-local paths, protect local conflicts, and
+install the validated session manifest after other restored durable files. The
+physical workflow and recovery procedures belong in
+[Operations](../operations.md).
+
+Restore and runner transitions for one session use the same local lock. A
+durable incomplete-restore marker blocks runner reuse after a partial restore;
+safe retry, rather than rollback of arbitrary local effects, is the recovery
+mechanism. Restored manifest-local references must be confined to the selected
+local session root, never trusted as producer-machine absolute paths.
+
+For the immutable remote-commit protocol, a restore or status operation binds
+to one pointer-selected commit and only its declared object identities. A force
+flag may replace an eligible regular managed file, but never turns a directory
+or other non-regular conflict into a successful restore.
+
+## Configuration
+
+Configuration is strict, explicit, centralized, and operator-oriented.
+
+- YAML decoding rejects unknown fields.
+- Defaults are centralized and testable.
+- Empty configured values do not silently replace meaningful defaults.
+- Validation rejects invalid composition before stage execution where
+  practical.
+- Session templating remains narrow and deterministic rather than becoming a
+  general configuration language.
+- Secret values are supplied indirectly and are not persisted in ordinary
+  configuration.
+
+Narratio must not become a second configuration system for downstream tools.
+External systems own their runtime defaults wherever practical; Narratio passes
+the paths required by its stage contracts and explicit operator overrides. The
+field-level contract and credential-supply mechanisms belong in
+[Configuration](../config.md).
+
+## Artifacts, Paths, And Storage
+
+Artifact identities and local and remote paths are application contracts.
+Canonical helpers own workspace, spool, cache, session, run, input, transcript,
+artifact, log, report, configuration, and publish-current paths. Callers must
+not reconstruct canonical paths through scattered string concatenation.
+
+Reusable audio cache entries require a typed record that binds a confined,
+no-follow regular file and its digest to the selected remote object identity.
+Size alone and unqualified multipart ETags are not content-integrity evidence.
+
+Artifact resolution is deterministic and manifest-aware. Producers materialize
+canonical outputs before reporting success, and consumers resolve declared
+artifact identities rather than infer files from unrelated directory contents.
+External artifact bundles become current only through validated immutable
+promotion and manifest records; directory presence alone never establishes
+availability.
+
+Writes, moves, replacements, and deletions must use narrow, explicit,
+root-confined destinations. Symlinks, traversal, broad roots, and ambiguous
+relative destinations must not expand the scope of an operation. Cleanup is
+permitted only through explicit operator action or configured post-publish
+gates, and it must preserve durable cache unless cache removal is explicitly
+requested.
+
+Physical layout, retention, and operational lifecycle belong in
+[Operations](../operations.md). Logical external formats and durable integration
+contracts belong under [Integrations](../integrations/).
+
+## Publish Commit Boundary
+
+Publish has one explicit remote commit boundary. A remote run becomes current
+only after Narratio has successfully uploaded its immutable run-scoped objects,
+the immutable commit manifest, and finally the current commit pointer.
+
+`current/commit-pointer.json` is the sole mutable selector and must be written
+exactly once, last. Failed, incomplete, skipped, or uncommitted publish attempts
+must not be presented as current remote state. Publish locks remain authoritative
+and are not bypassed by a forced run. Mutable remote locks use provider-enforced
+generation preconditions and are revalidated immediately before pointer
+selection; loss of that check leaves the prior committed snapshot current.
+
+Automatic local cleanup is permitted only after a successful publish commit,
+only when explicitly configured, and only through the path-safety guardrails.
+It is a durable local obligation bound to that committed run and its exact
+targets, not an inferred side effect of the current stage list. A cleanup
+failure makes the invocation incomplete while leaving the committed remote
+snapshot authoritative; later invocations resume the recorded obligation.
+
+## Security, Privacy, And Diagnostics
+
+Narratio distinguishes ordinary workspace data from credentials. Campaign and
+session material—including manifests, transcripts, prompts, generated
+configuration, logs, reports, diagnostics, and Notarius artifacts—is
+intentionally shareable with the configured workspace group. API-key material
+is sensitive and is not covered by the ordinary workspace-sharing policy.
+
+On POSIX systems, Narratio-created ordinary workspace directories converge on
+setgid `02775` and ordinary workspace files on `0664`, even when the caller's
+umask is restrictive. This preserves the existing workspace group for nested
+creation and atomic replacements without changing ownership. API-key storage
+uses a separate restrictive contract. On Windows, POSIX mode bits and setgid
+are not authoritative; operators must provide the equivalent shared-group and
+credential-restricted ACLs described in [Operations](../operations.md#workspace-permissions).
+
+Raw secrets must not be stored in pipeline, campaign, or session YAML or written
+to manifests, logs, generated configuration, reports, publish metadata,
+documentation, or examples. Secrets enter through configured environment
+variable names or secret-file references. Diagnostics should avoid transcript
+and prompt content unless a deliberate, bounded inspection mechanism requires
+it.
+
+Logs, reports, generated invocation files, generated configuration, and render
+debug files are diagnostics, not canonical pipeline products. They should be
+durable and discoverable where configured, and manifest references must preserve
+the distinction between diagnostics and artifacts.
+
+Documentation security rules belong in the
+[Documentation Policy](documentation.md). Credential supply belongs in
+[Configuration](../config.md), while permissions, sensitive runtime-artifact
+handling, and recovery belong in [Operations](../operations.md).
+
+## Determinism And Testability
+
+Narratio prefers deterministic behavior where practical, including stable local
+and remote layouts, sorted operation order, predictable generated
+configuration, repeatable command construction, deterministic artifact
+resolution, and reproducible planning.
+
+Run IDs and timestamps may be intentionally variable, but surrounding behavior
+must remain controllable in tests. Core behavior should be testable without live
+external services; expensive, nondeterministic, destructive, or external
+boundaries should be replaceable with focused test doubles. General testing
+philosophy and sufficiency rules belong in the [Testing Policy](testing.md).
+
+## Documentation And Decision Records
+
+Documentation follows the [Documentation Policy](documentation.md). Current
+behavior belongs in its canonical user, operator, integration, architecture, or
+internal owner. Proposed behavior and implementation status belong under
+`docs/roadmap/`.
+
+Significant architectural decisions may be recorded under `docs/adr/` using the
+format and lifecycle defined by the documentation policy. ADR acceptance does
+not establish that a decision has been implemented.
+
+## Architectural Non-Goals
+
+Narratio does not aim to provide:
+
+- a generic DAG or workflow engine;
+- a replacement configuration layer for WhisperX, Seriatim, Audita,
+  Scriptorium, or other downstream tools;
+- a storage abstraction broader than the needs of this pipeline;
+- stage logic coupled directly to cloud SDKs, transports, subprocess details,
+  or downstream implementation internals;
+- raw-secret persistence;
+- implicit cross-stage behavior that bypasses manifest and artifact contracts;
+  or
+- a prompt-authoring system.

docs/policy/documentation.md

@@ -0,0 +1,148 @@
+# Documentation Policy
+
+## Purpose
+
+This policy assigns each documentation topic to one canonical owner. Its goal is
+to keep Narratio documentation accurate, concise, discoverable, and resistant
+to drift for users, operators, developers, integrators, and LLM coding agents.
+
+## Core Rules
+
+### One Canonical Owner
+
+Each authoritative fact belongs in one document. A non-owning document may give
+a short, stable summary for orientation, but it must link to the canonical owner
+instead of repeating volatile details.
+
+Volatile details include commands, flags, configuration fields and defaults,
+stage or integration keys, schemas, file names, paths, status codes, retry
+behavior, and runtime guarantees. If readers could reasonably treat a statement
+as a contract, maintain it only in the owning document.
+
+### Current And Future Behavior
+
+Outside `docs/roadmap/`, documentation describes implemented behavior only.
+Partial features may be described only to their implemented boundary.
+
+ADRs are the narrow exception: an ADR may record an accepted architectural
+decision before implementation, but acceptance must not be presented as proof
+that the behavior exists. The roadmap owns implementation status and sequencing
+until the decision is implemented. Current architecture, user, operator,
+integration, and internal documentation are updated when the behavior lands.
+
+### Audience And Detail
+
+Write for the document's stated audience and include only the detail needed for
+its owned topic. User and operator docs should not expose implementation detail.
+Developer docs should link to user-facing and external contracts rather than
+restate them.
+
+### Examples
+
+Complete copyable files belong in `examples/`. Documentation may use the
+smallest illustrative snippet needed to explain its owned topic, but should link
+to maintained examples instead of embedding a second complete copy.
+
+Examples must be valid, secret-free, and tested where practical. Commands and
+configuration used in documentation should match the application.
+
+### Security And Privacy
+
+Documentation and examples must not contain real credentials, private keys,
+private environment dumps, sensitive source material, or private infrastructure
+details unless intentionally public. Document secret-handling mechanisms, not
+secret values.
+
+## Canonical Ownership
+
+| Topic | Canonical owner | Owned content | Content owned elsewhere |
+| --- | --- | --- | --- |
+| Product orientation and minimal end-to-end quickstart | `README.md` | What Narratio is, why it is useful, one shortest successful invocation, and links onward. | Complete command reference, configuration reference, operational procedures, implementation detail. |
+| Contributor entry point | `docs/development.md` | Task-oriented reading guide, minimal contributor orientation, baseline validation commands, and links to canonical docs. | Package inventory, architecture rules, subsystem behavior, detailed change recipes. |
+| Current application architecture | `docs/policy/architecture.md` | System shape, normative ownership, dependency direction, architectural boundaries, invariants, safety properties, and non-goals. | Concrete package inventory, implementation mechanics, contributor procedures, decision history, future work. |
+| Documentation organization | `docs/policy/documentation.md` | Documentation ownership, audience boundaries, maintenance rules, and ADR/document lifecycle. | Application architecture or product behavior. |
+| Testing policy | `docs/policy/testing.md` | Test philosophy, risk-based sufficiency, test boundaries, doubles, coverage guidance, regression-test policy, and criteria for adding, rewriting, or deleting tests. | Subsystem behavior, application contracts, subsystem-specific test inventories, and implementation plans. |
+| CLI contract | `docs/cli.md` | Commands, arguments, flags, invocation semantics, output conventions, and exit behavior. | End-to-end operating procedures, configuration field definitions, runtime filesystem layout, stage implementation details. |
+| Configuration contract | `docs/config.md` | Discovery and precedence, file schemas, fields, defaults, environment overrides, validation rules, and user-selectable stage or integration settings. | Complete example files, CLI syntax, runtime state lifecycle, implementation details. |
+| Operations | `docs/operations.md` | Runtime workflows, physical filesystem and remote-state layout, output and diagnostic handling, resume, cleanup, permissions, recovery, and operational limits. | CLI flag syntax, configuration field definitions, logical artifact schemas, implementation mechanics. |
+| Troubleshooting | `docs/troubleshooting.md` | Symptom-driven diagnosis, likely causes, safe inspection steps and remedies, and links to relevant contracts. | CLI syntax, configuration definitions, operational procedures, integration contracts, implementation mechanics. |
+| Public HTTP contract, if introduced | `docs/api.md` | Routes, authentication, media types, request and response schemas, status codes, pagination, caching, idempotency, rate limits, and HTTP retry semantics. | Client walkthroughs, upstream or downstream integration internals, implementation detail. |
+| Consumer guidance, if a public package or API is introduced | `docs/consumers/` | Task-oriented use of the public interface, minimal client examples, and consumer responsibilities. | HTTP wire semantics, external protocol contracts, internal implementation detail. |
+| External and durable integration contracts | `docs/integrations/` | External file formats and protocols, upstream and downstream contracts, logical artifact paths and schemas, media types, and compatibility behavior. | Physical runtime placement and lifecycle, internal transformations, CLI syntax, configuration defaults. |
+| Implemented component inventory | `docs/internal/overview.md` | Current packages and components, their implemented responsibilities, and links to focused internal docs. | Normative architecture, contributor reading policy, external contracts. |
+| Internal component behavior | Other files under `docs/internal/` | Implementation flow, internal collaborators and state transitions, package-local guarantees and failures, and relevant tests. | Global architecture invariants, configuration definitions and defaults, external schemas, operator procedures. |
+| Architectural decision history | `docs/adr/` | Significant decisions, context, alternatives, rationale, consequences, and supersession history. | Current behavior reference, implementation status, task sequencing. |
+| Future work and implementation status | `docs/roadmap/` | Proposed, accepted, deferred, or rejected work; implementation status; sequencing; and task breakdowns. | Implemented behavior reference and architectural decision rationale. |
+| Complete copyable artifacts | `examples/` | Maintained configuration, inputs, and other files intended to be copied or run. | Field-by-field reference, command reference, prose explanation. |
+
+Documents that do not exist are required only when the corresponding interface
+or responsibility exists. Do not create placeholder API, consumer, integration,
+or operations documents for behavior the application does not have.
+
+## Boundary Rules
+
+### Orientation
+
+The README owns product orientation. The developer guide routes contributors.
+Architecture owns normative structure. Internal overview owns the current
+concrete component map. These documents may link to one another but should not
+maintain parallel package or behavior descriptions.
+
+### Commands, Configuration, Operations, And Troubleshooting
+
+CLI documentation answers how to invoke the application. Configuration
+documentation answers what settings mean. Operations answers what happens to
+runtime state and how to operate or recover the application. Troubleshooting
+starts from observable symptoms and links readers to the owning command,
+configuration, operational, or integration contract. When a workflow crosses
+these topics, choose the document that owns the task and link to the other
+contracts.
+
+### Contracts And Implementation
+
+Integration and API documents define externally observable shapes and
+semantics. Internal documents explain how Narratio implements or consumes those
+contracts. Internal docs may name a field, file, or protocol to identify a
+dependency, but must link to its canonical contract for the definition.
+
+### Security Topics
+
+This policy owns what documentation and examples may contain. Architecture owns
+application security invariants. Configuration owns credential-supply
+mechanisms. Operations owns permissions and handling of sensitive runtime
+artifacts. Troubleshooting owns safe diagnostic and remediation guidance.
+Internal docs own implementation mechanisms only.
+
+## Architecture Decision Records
+
+Use sequentially numbered ADR filenames such as
+`0001-record-architecture-decisions.md`. Follow the lightweight Nygard format:
+
+1. title;
+2. status;
+3. date;
+4. context;
+5. decision;
+6. alternatives considered;
+7. consequences.
+
+Treat the decision content of an accepted ADR as immutable. When a decision
+changes, create a new ADR and update the earlier ADR's status to superseded.
+Rejected architectural alternatives belong in the ADR; rejected product ideas
+belong in the roadmap.
+
+## Maintenance
+
+When behavior changes, update its canonical owner in the same change. If
+ownership moves, remove the old definition and replace it with a link where
+navigation remains useful.
+
+Before completing documentation work:
+
+- verify affected behavior and examples;
+- check commands, flags, fields, defaults, schemas, and paths against their
+  implementation;
+- keep unimplemented behavior in the roadmap, subject to the ADR exception;
+- remove stale references and validate links;
+- confirm that non-owning documents summarize and link rather than redefine;
+- confirm that no secrets or sensitive private data were added.

docs/policy/testing.md

@@ -0,0 +1,296 @@
+# Testing Policy
+
+## Purpose
+
+Our tests exist to make **incorrect changes expensive and correct changes cheap**.
+
+We do not optimize for test count, line coverage, exhaustive isolation, or the fewest possible tests. We optimize for sufficient confidence in important behavior while imposing as little unnecessary friction as possible on future development.
+
+## Every test has a cost
+
+Testing is not an unqualified good. Every test imposes both an immediate cost and a continuing lifetime cost.
+
+A test must be:
+
+- written and reviewed;
+- understood by future maintainers and coding agents;
+- executed in local and CI workflows;
+- diagnosed when it fails;
+- updated when legitimate behavior changes;
+- maintained as fixtures, APIs, and dependencies evolve; and
+- removed or rewritten when it becomes redundant, brittle, misleading, or obsolete.
+
+Tests also create cognitive and architectural friction. They can constrain refactoring, duplicate policy, slow feedback loops, add noise to failures, and cause harmless implementation changes to require unrelated edits across the suite.
+
+A test is warranted only when the confidence it provides justifies these costs.
+
+Apply this cost-benefit analysis at two levels:
+
+1. **Per test:** What realistic defect does this test detect, how consequential would that defect be, and is that protection worth the test's lifetime cost?
+2. **Across the suite:** Does this collection provide materially more confidence than a smaller, simpler suite would?
+
+The preferred test suite is a **lean suite that provides sufficient confidence in the risks that matter, without redundant or low-value tests**. We seek sufficient confidence with the least unnecessary testing friction, not the fewest possible tests.
+
+Some friction is intentional. Tests should make dangerous changes—such as breaking compatibility, corrupting data, violating security boundaries, or reintroducing subtle bugs—require deliberate review. They should not make ordinary internal changes needlessly expensive.
+
+The cost of a test is not a reason to omit testing by default. Do not cite maintenance cost abstractly. When omitting a plausible test, be able to state why the protected failure is low-risk, already covered, obvious, reversible, or cheaper to detect elsewhere. For consequential, subtle, or difficult-to-observe behavior, the presumption should favor testing.
+
+## Default testing style
+
+Use a **classical/Detroit-style** approach:
+
+- Test observable behavior, resulting state, contracts, and invariants.
+- Use real internal collaborators when they are fast and deterministic.
+- Use fakes, stubs, or mocks primarily at expensive, nondeterministic, destructive, or external boundaries.
+- Prefer package-level behavioral tests over tests coupled to private helpers or internal call sequences.
+- Treat exact collaborator interactions as testable behavior only when the interaction itself is a requirement.
+
+Examples of appropriate seams include clocks, randomness, subprocesses, remote APIs, object storage, email, and paid LLM calls.
+
+## Test execution requirements
+
+Tests in the default suite must be deterministic, offline, and independent of real credentials. They must not invoke paid APIs or depend on mutable external services. Tests that require live infrastructure must be explicitly opt-in and clearly separated from the default suite.
+
+Control clocks, randomness, environment variables, and other process-global or machine-specific state when they affect behavior. Tests should be safe to run repeatedly and alongside other tests without depending on execution order or state left by an earlier test.
+
+## What deserves tests
+
+Prioritize tests for:
+
+1. Public and package-level contracts.
+2. Domain rules and important invariants.
+3. Boundary conditions and malformed input.
+4. Failure handling, cancellation, retries, recovery, and partial success.
+5. Serialization, schemas, compatibility, and round trips.
+6. Previously observed or plausible regressions.
+7. Representative integration and end-to-end workflows.
+
+A package-level contract is behavior relied upon by another package or major collaborator, not every observable detail of a package implementation.
+
+For behavior involving **data integrity, destructive operations, compatibility, security, concurrency, idempotency, or recovery**, presume that durable tests are required unless the behavior is already credibly protected at another layer.
+
+Do not add tests merely because a function, branch, or line exists. Do not add a test when the same meaningful risk is already adequately protected elsewhere.
+
+## Choose the right test boundary
+
+Test through the narrowest stable boundary that expresses the behavior clearly.
+
+This is often the package API, but it may instead be:
+
+- a smaller pure function when dense domain logic is most clearly isolated there;
+- a package-level operation when several internal collaborators jointly produce the behavior; or
+- a larger integration boundary when correctness emerges from interaction with a real dependency.
+
+Do not force all behavior through oversized end-to-end tests. Do not test every private helper merely because it exists. Choose the boundary that gives durable confidence with the least incidental coupling.
+
+## Test behavior, not implementation
+
+A test should protect a decision, contract, or invariant—not memorialize the current implementation.
+
+Before adding or retaining a test, ask:
+
+> What realistic defect would this test catch?
+
+A test is suspect when its main purpose is to detect that someone:
+
+- changed an internal constant;
+- renamed or split a private helper;
+- reordered equivalent internal operations;
+- changed incidental formatting;
+- replaced one correct algorithm with another; or
+- refactored internal object structure without changing behavior.
+
+Refactoring should normally require no test edits unless the refactored structure is itself part of the contract.
+
+A test can be factually correct and still have negative value. Accurately describing current behavior is not enough; the protected behavior must be important enough to justify the future friction.
+
+## Expected effects of different changes
+
+Use the following expectations when evaluating test failures and test maintenance:
+
+| Change | Expected effect on tests |
+|---|---|
+| Internal refactor that preserves behavior | Existing tests should normally remain unchanged and continue to pass. |
+| Change to an internal default with no contractual significance | Behavioral tests should normally remain unchanged; tests should derive expectations from configuration or relationships rather than duplicate the old value. |
+| Intentional change to public behavior, policy, schema, or compatibility guarantees | The relevant tests should be reviewed and changed deliberately. |
+| Accidental violation of a contract or invariant | Tests should fail; fix the production code rather than rewriting the tests to accept the defect. |
+
+A test failing is not the same as a test needing to be edited. Many tests may correctly fail because of one production defect. The maintenance smell is a correct internal change that requires unrelated expectation updates throughout the suite.
+
+## Separate mechanism from policy
+
+Configurable thresholds and defaults must not be duplicated throughout the test suite.
+
+For example, do not encode an internal concurrency limit indirectly:
+
+```go
+// Production policy:
+const maxConcurrency = 4
+
+// Brittle test:
+err := startProcesses(5)
+require.Error(t, err)
+```
+
+Instead, test the mechanism relationally:
+
+```go
+const limit = 2
+runner := NewRunner(limit)
+
+require.NoError(t, runner.Start(limit))
+require.ErrorIs(t, runner.Start(limit+1), ErrTooMuchConcurrency)
+```
+
+The test should prove:
+
+- the configured limit is accepted; and
+- one beyond the configured limit is rejected.
+
+The production default should be tested exactly only when its literal value is itself a public, operational, safety, protocol, or compatibility requirement.
+
+Apply the same rule to limits, timeouts, capacities, retry counts, and ranges: test relationships and behavior, not duplicated literals.
+
+For concurrency limits, test both kinds of behavior when relevant:
+
+1. **Configuration enforcement:** invalid or excessive requested values are handled correctly.
+2. **Runtime enforcement:** observed peak concurrency never exceeds the configured limit.
+
+Use a test-controlled limit and measure the behavior relative to that limit. Do not merely assert today's default value.
+
+## Avoid semantic duplication across layers
+
+Each behavior should have a clear test owner.
+
+- Parser tests own parsing cases.
+- Validator tests own validation rules.
+- Domain tests own transformations and invariants.
+- Adapter tests own external integration behavior.
+- Orchestrator tests own coordination and failure propagation.
+- CLI tests own argument and configuration mapping.
+- End-to-end tests prove that representative assembled workflows work.
+
+Higher-level tests should not repeat every lower-level case. A single intentional policy change should not require unrelated edits across many test files.
+
+Tests that are individually reasonable may still be collectively redundant. Evaluate the marginal value of each additional test in light of the protection already provided by the rest of the suite.
+
+## Use test doubles deliberately
+
+Choose the least elaborate test double that provides the required control or observation.
+
+As a default:
+
+1. Prefer real collaborators when they are fast and deterministic.
+2. Use small in-memory fakes when realistic stateful behavior is helpful.
+3. Use stubs when a dependency only needs to provide controlled responses.
+4. Use mocks when the interaction itself is contractual.
+
+Mocks are appropriate when the contract includes facts such as:
+
+- a notification is sent exactly once;
+- a transaction is committed only after successful writes;
+- cancellation reaches a subprocess;
+- an expensive API is called no more than once; or
+- a security audit event is emitted.
+
+Do not use mocks merely to isolate every object or reproduce the implementation's call graph.
+
+## Go-specific guidance
+
+Use:
+
+- table-driven tests for meaningful behavioral categories and boundaries;
+- `t.TempDir()` for real filesystem behavior;
+- `httptest.Server` for realistic HTTP interactions;
+- fuzz tests for parsers, normalization, path handling, and broad input spaces;
+- golden files only when the complete output is intentionally stable;
+- integration tests where correctness depends on component interaction; and
+- a small number of representative end-to-end tests.
+
+Avoid exact error-string assertions unless the wording is itself contractual. Prefer `errors.Is`, `errors.As`, typed errors, or structured error fields.
+
+At CLI boundaries, prefer exit classifications, structured output, and the smallest stable semantic fragment needed to identify the error. Do not snapshot complete diagnostic wording unless it is contractual.
+
+Golden-file updates must require an explicit local flag. CI must not update golden files automatically, and reviewers must inspect the semantic diff before accepting an update.
+
+Keep tests readable and direct. Test helpers and fixture frameworks must earn their own maintenance cost; do not build elaborate test infrastructure for small or isolated needs.
+
+## Coverage
+
+Coverage is a diagnostic, not a target.
+
+Use it to find untested critical branches and unexpectedly weak packages. Do not write low-value tests solely to increase a percentage, and do not infer test quality from coverage alone.
+
+Pure domain logic will often warrant higher coverage than CLI wiring or external adapters. Uneven coverage is acceptable when it reflects risk.
+
+Increasing coverage is valuable only when the newly covered behavior protects a meaningful risk at an acceptable cost.
+
+## Regression tests
+
+A bug fix should normally include a regression test that fails before the fix and passes afterward.
+
+Retain the test when the defect could realistically recur and its consequences justify the ongoing cost. Prefer the narrowest durable test of the violated contract or invariant; do not preserve accidental implementation details from the original bug.
+
+Not every historical bug requires a permanent test. If the underlying design has made recurrence impossible, the test has become redundant, or a stronger invariant test now subsumes it, remove or consolidate it.
+
+## Deleting or rewriting tests
+
+Tests are maintained code, not permanent historical artifacts.
+
+Delete or rewrite a test when its maintenance cost exceeds the confidence it provides.
+
+Strong candidates include tests that:
+
+- require updates after harmless internal changes;
+- directly assert private constants without protecting a real contract;
+- duplicate the same policy across several layers;
+- verify mock choreography rather than outcomes;
+- snapshot large amounts of incidental output;
+- test trivial private helpers already exercised through stable package behavior;
+- protect risks already covered more effectively elsewhere;
+- are flaky, misleading, obsolete, or disproportionately expensive to diagnose; or
+- no longer correspond to a plausible failure mode.
+
+Several brittle tests may encode one genuine requirement. Replace them with one durable behavior-level or invariant test rather than preserving all of them.
+
+Deleting a low-value test can improve the quality of the suite by reducing noise, maintenance burden, and friction around legitimate change.
+
+## Reviewing a proposed test
+
+Use the following questions when the value, boundary, or durability of a proposed test is not self-evident. Significant test additions should be reviewable against them, but written answers are not required for every routine test.
+
+1. What realistic defect would it catch?
+2. How likely is that defect?
+3. How consequential would it be?
+4. Is the behavior already protected elsewhere?
+5. At which layer should this behavior be owned?
+6. Does the test assert a durable contract or an incidental implementation detail?
+7. Could the implementation be refactored without changing the behavior and without editing this test?
+8. What should cause this test to fail?
+9. What legitimate changes should not cause this test to fail?
+10. What ongoing maintenance, execution, and diagnostic cost will the test impose?
+11. Is there a smaller or more direct test that protects the same risk?
+
+Do not add the test when its expected lifetime cost exceeds its expected protective value.
+
+When deciding not to test plausible behavior, record or be able to explain why the risk is low, already protected, obvious, reversible, or cheaper to detect elsewhere.
+
+## Definition of sufficient
+
+A test suite is sufficient when:
+
+- important contracts and invariants are protected;
+- meaningful boundaries and failure modes are exercised;
+- realistic and consequential regressions are credibly protected against silent recurrence;
+- behavior involving data integrity, destructive operations, compatibility, security, concurrency, idempotency, and recovery is credibly protected;
+- important external boundaries have realistic integration coverage;
+- representative complete workflows are tested;
+- failures provide useful signal rather than redundant noise;
+- legitimate internal changes usually do not require test edits; and
+- additional tests would mostly repeat existing protection or preserve inconsequential implementation details.
+
+Sufficiency is a risk judgment, not a coverage percentage or test count. Reassess it as the application, its users, and the consequences of failure evolve.
+
+The governing rule is:
+
+> Test heavily where failure is consequential, subtle, or difficult to detect after the fact. Test lightly where failure is obvious, reversible, and inexpensive—and retain no test whose lifetime cost exceeds the confidence it provides.

docs/roadmap/implementation.md

@@ -0,0 +1,546 @@
+# Notarius v0.6 CLI References Implementation Plan
+
+## Purpose And Status
+
+This is the executable implementation plan for the accepted target state in
+[Notarius v0.6 CLI Reference Integration](notarius-v0.6-cli-references.md). It
+is written for a `gpt-5.6-terra` coding agent that will implement exactly one
+pending stage per prompt, in order.
+
+The feature roadmap owns user intent, architectural boundaries, settled policy,
+and the target end state. This document owns delivery order, concrete changes,
+test allocation, and implementation status. Do not restate or change a roadmap
+decision here during implementation; if current Notarius v0.6.0 evidence
+contradicts the roadmap, stop and record the conflict instead of inventing a
+different contract.
+
+| Stage | Outcome | Status |
+| ---: | --- | --- |
+| 1 | Add the reference-selector and configuration vocabulary, including optional spell-catalog inputs. | Completed |
+| 2 | Materialize and inventory the optional spell catalog through the prepare and operator lifecycle. | Completed |
+| 3 | Centralize manifest-authoritative prepared-input resolution and migrate analyze to it. | Completed |
+| 4 | Add deterministic Notarius v0.6 reference arguments at the subprocess adapter boundary. | Completed |
+| 5 | Resolve references in extract and bind fingerprints, resume, and metadata to their identities. | Completed |
+| 6 | Prove assembled extraction lifecycle and downstream invalidation behavior. | Completed |
+| 7 | Update canonical documentation and maintained examples for the completed feature. | Completed |
+| 8 | Perform compatibility, quality, and repository-wide closure validation. | Completed |
+
+## Governing Decisions
+
+The following requirements are settled and are not questions for the
+implementing agent:
+
+1. `pipeline.notarius.references` is a map whose key is a Notarius v0.6 CLI
+   reference selector and whose value is a prepared Narratio source ID. It is
+   not a map of paths.
+2. Every configured binding is required. There is no per-entry `required`
+   field. An optional Notarius reference is omitted by omitting the map entry.
+   An empty or omitted map remains valid for custom pipelines and backward
+   compatibility.
+   The map is limited by the centrally declared configuration constant
+   `MaxNotariusReferenceBindings = 256`, which is far above the four-entry
+   maintained D&D case while bounding argv and manifest growth.
+3. The supported prepared reference sources are
+   `narratio.input.party`, `narratio.input.players`,
+   `narratio.input.glossary`, and `narratio.input.spell_catalog`.
+   Arbitrary Notarius slot names and qualified selectors may bind those sources;
+   direct paths and later-stage artifacts may not.
+4. Campaign and session `spell_catalog_file` are optional. A session value
+   overrides the campaign value; an empty session value inherits the campaign
+   value. A `narratio.input.spell_catalog` reference binding requires an
+   effective configured file.
+5. Prepared sources are authoritative only when the current session manifest
+   records the matching canonical input and checksum. Consumers do not fall
+   back to campaign/session source paths or accept an incidental workspace file.
+6. Narratio passes absolute prepared-file paths to Notarius. Fingerprints and
+   metadata use the canonical workspace-relative path identity together with
+   selector, source ID, SHA-256 checksum, and size so workspace relocation does
+   not become the only identity signal.
+7. External CLI references are sorted by normalized selector. They override
+   matching external references in the Notarius configuration. Narratio never
+   emits `--without-reference`, the deprecated `roster` alias, or CLI bindings
+   for generated D&D artifact handoffs.
+8. Notarius remains authoritative for whether a selected target declares a
+   slot, accepted reference media types and sizes, generated-handoff collisions,
+   pipeline topology, and D&D payload schemas. Narratio validates selector
+   structure and its own source contract only.
+9. Notarius v0.6.0 is the minimum supported CLI contract when references are
+   configured. Do not add version-string parsing or an automatic per-session
+   `notarius config validate` subprocess.
+10. The existing receipt-v2, bundle-confinement, diagnostic, ten-lane selection,
+    immutable promotion, and analysis-source behavior must remain intact.
+11. The default test suite remains offline, deterministic, and independent of
+    a sibling checkout or installed Notarius binary. A real v0.6.0 smoke run is
+    useful supplementary evidence when locally available, not a default-suite
+    dependency.
+12. Add no external Go dependency for this feature. Use narrow owner-specific
+    types and existing file, path, artifact, manifest, adapter, and stage
+    facilities.
+
+## Instructions For Every Stage
+
+For each implementation prompt, the coding agent must:
+
+1. Read `docs/development.md`, all three files under `docs/policy/`, the feature
+   roadmap, this plan, and the stage-specific documents and source named below.
+   Inspect the current tree because earlier stages may have changed names or
+   ownership boundaries.
+2. Use the repository knowledge graph first for code discovery and call tracing;
+   use text search for documentation, configuration, examples, string literals,
+   and evidence the graph cannot supply.
+3. Confirm the worktree state and preserve unrelated changes. Implement only the
+   current stage. Do not begin a later stage merely because an adjacent file is
+   open.
+4. Keep production code, focused tests, fakes, and fixtures consistent within
+   the stage. Remove superseded helpers when their final caller migrates. Do not
+   retain two competing source maps, path resolvers, fingerprint paths, or
+   subprocess argument builders.
+5. Follow the testing policy's ownership rule. Parser/config tests own selector
+   and configuration cases; artifact tests own prepared-file identity and
+   integrity; adapter tests own exact arguments; stage tests own orchestration and
+   resume; application tests own lifecycle invalidation. Do not repeat every
+   lower-level case at higher levels.
+6. Keep errors actionable and content-free. They may identify a selector,
+   Narratio source ID, canonical path, or checksum mismatch, but must not include
+   reference contents. Preserve ordinary group-workspace permissions and
+   restrictive API-key handling.
+7. Run `gofmt` on changed Go files and focused tests while iterating. Before
+   marking any stage complete, run at minimum:
+
+   ```sh
+   go test ./...
+   go test -race ./...
+   go vet ./...
+   go build ./...
+   go test ./internal/doccheck
+   go test ./internal/config -run '^TestExamplesLoadAndValidate$'
+   ```
+
+   Default tests must not contact live services or require credentials.
+8. Compare the final diff against the stage goal and exit criteria. Update only
+   the current stage's status row from `Pending` to `Completed`. Do not mark a
+   stage complete while a required check fails or required behavior is absent.
+   Intermediate commits are implementation-branch state and must not be released
+   before Stage 7 has reconciled current-behavior documentation.
+
+## Stage 1 — Reference And Configuration Vocabulary
+
+**Read first:** `docs/config.md`, `docs/integrations/notarius.md`,
+`internal/config/config.go`, `internal/config/defaults.go`,
+`internal/config/load.go`, `internal/config/validate.go`,
+`internal/config/notarius_test.go`, `internal/config/campaign_config_test.go`,
+and `internal/artifactpolicy/policy.go` and its tests. Read the tagged Notarius
+v0.6.0 `docs/cli.md` reference-selector section from `../notarius` when that
+checkout is available; otherwise use the canonical link from the feature
+roadmap.
+
+**Depends on:** None.
+
+**Goal:** Establish one normalized reference-selector grammar and the strict
+configuration model needed by later stages, without adding path-valued Notarius
+configuration or making spell catalogs mandatory for every campaign.
+
+**Work:**
+
+- Add a small dependency-free `internal/notariusref` package as the contract
+  owner for Notarius reference selector normalization. Its exported normalizer
+  must trim the selector and each component, reject empty components and `=`,
+  and accept only the v0.6 forms `slot`,
+  `chunk.slot`, `lane.slot`, `lane.extract.slot`, `lane.merge.slot`, and
+  `lane.normalize.slot`. A three-component selector accepts only `extract`,
+  `merge`, or `normalize` in its middle component. Do not check the selector
+  against a Notarius module or lane registry.
+- Add `References map[string]string` with YAML key `references` to
+  `config.NotariusConfig`. During enabled Notarius validation, sort raw keys,
+  normalize each selector through the shared contract, trim each source value,
+  reject empty values and normalized-selector collisions, require each value to
+  be one of the four prepared reference sources, and replace the config map with
+  its normalized form. Enforce the named, centrally discoverable
+  `config.MaxNotariusReferenceBindings` limit of 256 entries with an error that
+  identifies the field and limit. Keep a nil/empty map valid. Rely on strict YAML
+  decoding to reject duplicate identical keys, but explicitly reject distinct
+  raw keys that normalize to one selector.
+- Add `SpellCatalogFile string` with YAML key `spell_catalog_file` to campaign
+  inputs and session inputs, plus `SpellCatalogFile ResolvedInputFile` to
+  resolved stable inputs. Merge it with the existing session-over-campaign
+  helper. It is not part of the campaign-required input set. Reject a non-empty
+  configured scalar that becomes empty after trimming.
+- Add `artifactpolicy.SourceInputSpellCatalog` and make the artifact-policy
+  owner describe all four prepared reference sources, including their canonical
+  manifest kind and filename. Use that owner for stable-source recognition
+  instead of adding a second switch in configuration validation. Preserve the
+  existing three source IDs and their behavior.
+- Extend cross-configuration validation so a normalized reference to
+  `narratio.input.spell_catalog` requires a non-empty effective resolved
+  `spell_catalog_file`. Existing campaign requirements already guarantee party,
+  players, and glossary declarations. Do not check filesystem existence during
+  configuration validation.
+- If analyze's current private filename switch must change to keep the tree
+  behaviorally coherent, make it delegate to the artifact-policy descriptor and
+  recognize spell catalog; Stage 3 will replace the filesystem-only resolver.
+
+**Tests and exit criteria:** At the contract/config owners, cover every accepted
+selector shape; zero, empty, excess, invalid-stage, and `=` forms; whitespace
+normalization; normalized collisions; unsupported and empty source IDs; nil and
+empty maps; exactly the configured binding limit and limit plus one; strict
+unknown fields; campaign inheritance and session override; optional omission;
+and the cross-config missing-spell-catalog failure. Prefer table-driven parser
+and validator tests over assertions against private helper structure. Existing
+configuration and example tests must still pass without adding spell catalogs
+to every campaign. The codebase has one selector grammar owner and one
+prepared-source descriptor owner.
+
+## Stage 2 — Spell Catalog Prepare And Operator Lifecycle
+
+**Read first:** `docs/internal/stage-prepare.md`, `docs/internal/workspace.md`,
+`docs/operations.md`, `internal/stage/prepare.go` and its tests,
+`internal/app/operator_inspection.go`, `internal/app/operator_findings.go` and
+their focused tests, `internal/manifest/manifest.go`, and the relevant confined
+file-operation helpers.
+
+**Depends on:** Stage 1.
+
+**Goal:** Make the effective optional spell catalog a normal prepared session
+input with canonical storage, checksum/provenance, safe stale-file handling, and
+operator visibility.
+
+**Work:**
+
+- Resolve `StableInputs.SpellCatalogFile` with the same origin-preserving
+  campaign/session behavior as the five existing stable inputs. When configured,
+  require a regular readable source, copy it atomically to
+  `inputs/spell_catalog.json`, preserve ordinary workspace permissions, and add
+  one manifest input record with kind `spell_catalog`, canonical destination,
+  checksum, and `campaign_config` or `session_config` source provenance.
+- Treat the input as optional when no effective path is configured. Do not call
+  the required-input path resolver with an empty value and do not create a
+  manifest record. Remove an obsolete canonical `inputs/spell_catalog.json`
+  without following it when a forced prepare transitions from configured to
+  absent; refuse to recursively remove a directory or other ambiguous object at
+  that exact file path.
+- Include a configured spell catalog in operator inspection and validation
+  findings. Omission is not an error unless Stage 1 cross-configuration policy
+  says the enabled Notarius reference requires it. Reuse the common resolved
+  stable-input enumeration where practical instead of extending parallel
+  hand-written lists in several functions.
+- Adjust input-slice capacity, deterministic ordering, test fixtures, and any
+  manifest assumptions affected by the optional sixth stable file. Do not parse
+  or schema-validate the JSON payload in Narratio; Notarius owns that contract.
+
+**Tests and exit criteria:** Through prepare and operator package behavior, cover
+campaign and session source provenance, canonical destination bytes and
+checksum, optional omission, missing configured source, replacement after source
+change, safe removal when configuration is removed, rejection of an ambiguous
+destination object, deterministic manifest ordering, and operator reporting.
+Do not duplicate selector-validation cases from Stage 1. Existing sessions with
+no spell catalog remain valid and produce no stale manifest entry.
+
+## Stage 3 — Manifest-Authoritative Prepared Input Resolution
+
+**Read first:** `docs/internal/artifacts.md`, `docs/internal/manifest.md`,
+`docs/internal/stage-analyze.md`, `internal/artifacts/artifact_resolver.go`,
+`internal/artifacts/resolve.go`, `internal/artifacts/checksum.go`, their tests,
+and the prepared stable-input resolution path in `internal/stage/analyze.go` and
+`internal/stage/analyze_test.go`.
+
+**Depends on:** Stage 2.
+
+**Goal:** Give analyze and extract one integrity-checked resolver for prepared
+stable sources so neither stage trusts incidental files or reconstructs its own
+source-to-filename table.
+
+**Work:**
+
+- Add an artifacts-owned `PreparedInputIdentity` contract containing source ID,
+  manifest kind, absolute canonical path, slash-separated path relative to the
+  session root, SHA-256 checksum, and byte size. Add one resolver that accepts
+  session paths, the current session manifest, and a stable source ID. Provide a
+  typed or sentinel absence classification so callers can distinguish no current
+  manifest record from corrupt or unsafe recorded evidence.
+- Derive kind and filename exclusively from the artifact-policy descriptor. The
+  resolver must require exactly one current manifest input record with the
+  expected kind and canonical path; resolve/rebase recorded local paths through
+  existing session-local path safety helpers; require the result to equal the
+  canonical file below `inputs/`; reject escapes, symlinks, non-regular files,
+  empty files, missing checksums, duplicate records, and checksum mismatches; and
+  calculate size without loading the complete file into memory. Do not fall back
+  to the configured campaign/session path or accept canonical file presence
+  without manifest evidence. Zero matching manifest records is the typed absent
+  case; once a record exists, a missing or invalid file is an integrity error,
+  not optional absence.
+- Return owner-neutral errors from `internal/artifacts`. At stage boundaries,
+  wrap unavailable or stale prepared inputs with the source ID and actionable
+  `narratio run-stage prepare <session_id> --force` guidance. Do not include file
+  contents.
+- Replace analyze's private prepared-source filename switch and filesystem-only
+  resolver with the shared artifact resolver. Preserve required-versus-optional
+  Scriptorium input behavior: a typed absent optional source is omitted, an
+  absent required source fails with prepare guidance, and invalid recorded
+  evidence fails regardless of optionality. Make `narratio.input.spell_catalog`
+  usable wherever another prepared Scriptorium source is accepted.
+
+**Tests and exit criteria:** Artifact-package tests own valid resolution and the
+missing-record, duplicate-record, wrong-kind/path, traversal/rebase, symlink,
+non-regular, empty, missing-checksum, and checksum-mismatch boundaries. Analyze
+tests need only prove required/optional stage behavior and successful use of the
+shared source, including spell catalog; do not repeat the artifact resolver's
+full matrix. Remove the old filename/path resolver after its final caller moves.
+
+## Stage 4 — Notarius Adapter Reference Arguments
+
+**Read first:** `docs/internal/adapters.md`, `docs/integrations/notarius.md`,
+`internal/adapters/notarius/runner.go`, `fake.go`, `subprocess.go`, and focused
+adapter tests. Re-read the Notarius v0.6.0 subprocess and CLI reference-selector
+contracts from the tagged sibling checkout when available.
+
+**Depends on:** Stage 1.
+
+**Goal:** Extend the transport-neutral Notarius request and production adapter
+to emit safe, exact, repeatable v0.6 `--reference` arguments without changing
+receipt or bundle ingestion.
+
+**Work:**
+
+- Add a transport-neutral reference binding containing normalized selector and
+  absolute path, and add an ordered slice of those bindings to `RunRequest`.
+  Keep source IDs and manifest identities out of the adapter contract; those are
+  stage policy.
+- Validate each adapter binding before process launch: normalize/validate the
+  selector through the shared Stage 1 contract, require a non-empty absolute
+  path, reject duplicate normalized selectors, and avoid mutating the caller's
+  slice. Do not open or parse the reference file in the adapter.
+- Build arguments as repeated pairs `--reference`,
+  `<normalized-selector>=<absolute-path>` after `--output-dir` and before
+  `--json`. Preserve one argument for the combined selector/path value so spaces,
+  additional `=` characters within the path portion, and platform separators do
+  not involve shell interpretation. The request order is authoritative; Stage 5
+  will supply sorted bindings.
+- Preserve current executable, environment, timeout, cancellation, diagnostic,
+  receipt-v2, bounded-read, confinement, and bundle-discovery behavior. Do not
+  add `--without-reference`, generated reference arguments, version probing, or
+  configuration preflight.
+- Update the fake only as required to retain and expose the extended request.
+
+**Tests and exit criteria:** Adapter tests own exact argv with zero and multiple
+references, position before `--json`, spaces and `=` in paths, selector
+normalization, duplicate/invalid selector rejection, relative/empty path
+rejection, and no subprocess start after request-validation failure. Existing
+receipt-v2 and bundle fixture tests must remain unchanged in meaning and pass.
+Do not assert stage-level source sorting here beyond preserving the request
+order.
+
+## Stage 5 — Extract Reference Identity, Invocation, And Resume
+
+**Read first:** `docs/internal/stage-extract.md`,
+`docs/integrations/notarius.md`, `docs/internal/manifest.md`,
+`internal/stage/extract.go`, `internal/stage/extract_resume.go`, their focused
+tests, the Stage 3 prepared-input identity contract, and the Stage 4 Notarius
+request contract.
+
+**Depends on:** Stages 3 and 4.
+
+**Goal:** Make configured references part of the actual extraction invocation
+and durable reuse contract, using one resolution path for initial execution and
+resume validation.
+
+**Work:**
+
+- Add one extract-owned reference-resolution helper used by both `Run` and
+  `ValidateResume`. Iterate normalized config bindings in lexical selector
+  order, resolve each source through the Stage 3 manifest-authoritative resolver,
+  and produce both adapter bindings and immutable reference identities. Resolve
+  every reference before creating run-local receipt, log, output, or promotion
+  directories and before invoking the adapter.
+- Define the fingerprint/metadata identity as normalized selector, source ID,
+  canonical session-relative slash path, SHA-256 checksum, and byte size. Do not
+  include contents or original campaign/session absolute paths. Pass only
+  selector and absolute prepared path to the adapter.
+- Extend the extraction fingerprint document with the sorted reference
+  identities. Keep all existing binary, config path, pipeline, timeout, working
+  directory, trimmed-transcript identity, and required-output identities. The
+  result must be independent of YAML map iteration order and must change for a
+  selector, source, relative path, checksum, or size change.
+- Persist `reference_count` and a bounded deterministic `references` metadata
+  list on successful extraction. Each entry contains exactly `selector`,
+  `source_id`, `path`, `checksum`, and `size_bytes`. Empty bindings produce count
+  zero and an empty list. Do not duplicate Notarius reference payloads or
+  downstream error messages.
+- Make resume recompute current reference identities through the same helper
+  before comparing the configuration fingerprint. A valid changed prepared
+  input yields a fingerprint mismatch and a non-resumable result so extraction
+  reruns. Missing, unsafe, or checksum-inconsistent current input is an error
+  with prepare-force guidance because immediately rerunning extract cannot
+  succeed. Do not silently reuse the old bundle.
+- Preserve explicit disabled-stage skip without resolving references. Preserve
+  required lane selection, immutable promotion, receipt identity, and bundle
+  evidence behavior.
+
+**Tests and exit criteria:** Stage tests own sorted request construction for all
+four D&D bindings, zero bindings, failure before adapter invocation for an
+unavailable source, content-free metadata, and fingerprint changes for each
+identity field while remaining stable across map order. Resume tests must prove
+reuse with unchanged references, non-reuse after a valid prepared-reference
+change, hard failure for missing or checksum-invalid current evidence, and no
+reference resolution when disabled. Use the fake adapter; do not duplicate exact
+subprocess argv cases from Stage 4.
+
+## Stage 6 — Assembled Lifecycle And Invalidation Coverage
+
+**Read first:** `docs/internal/overview.md`, `docs/internal/manifest.md`,
+`docs/internal/stage-extract.md`, `docs/internal/stage-prepare.md`,
+`internal/app/runner.go`,
+`internal/app/extract_lifecycle_test.go`, and representative full pipeline and
+stage fixtures. Inspect existing downstream invalidation tests before adding
+new cases.
+
+**Depends on:** Stage 5.
+
+**Goal:** Prove at the application boundary that prepared campaign context
+reaches Notarius and that reference changes cannot leave extraction or later
+analysis falsely current.
+
+**Work:**
+
+- Extend the smallest existing assembled runner fixture to execute prepare and
+  extract with party, players, glossary, and spell catalog bindings. Assert that
+  the fake Notarius request receives the four canonical prepared absolute paths,
+  not the original campaign/session source paths, and that the successful
+  manifest records bounded reference identity.
+- Add one lifecycle regression covering a valid reference-content change:
+  rerun/force prepare so the manifest and prepared checksum change, then verify
+  extract resume is rejected, Notarius runs again, and succeeded canonical
+  downstream stages are invalidated according to the existing stage-order
+  policy. Assert outcomes, not private runner call choreography.
+- Add one representative session override case to prove the overridden prepared
+  bytes/checksum reach extraction. Do not repeat all four configuration merge
+  cases or artifact-integrity failures already owned by earlier stages.
+- Confirm an empty reference map preserves the pre-v0.6 invocation behavior and
+  that all ten configured D&D lanes remain registered as the same
+  `narratio.extraction.<key>` sources available to analyze.
+- Fix production integration defects exposed by these assembled tests without
+  broadening the feature or adding a DAG, generic reference workflow, or direct
+  Notarius payload parsing.
+
+**Tests and exit criteria:** The application-level tests must be deterministic,
+offline, and fake only the external Notarius boundary. They must credibly fail
+if Narratio passes original paths, omits one configured reference, reuses stale
+extraction, or loses a configured lane, while remaining insensitive to private
+helper structure and exact non-contractual diagnostics. Earlier focused suites
+and the repository baseline remain green.
+
+## Stage 7 — Canonical Documentation And Maintained Examples
+
+**Read first:** `docs/policy/documentation.md`, `docs/config.md`,
+`docs/operations.md`, `docs/troubleshooting.md`,
+`docs/integrations/notarius.md`, `docs/internal/overview.md`,
+`docs/internal/adapters.md`, `docs/internal/artifacts.md`,
+`docs/internal/stage-prepare.md`, `docs/internal/stage-extract.md`,
+`docs/internal/stage-analyze.md`, `examples/README.md`, and all maintained
+pipeline, campaign, and session examples affected by the new fields.
+
+**Depends on:** Stage 6.
+
+**Goal:** Move the completed behavior from roadmap-only future state into its
+canonical current-behavior owners and provide valid copyable D&D examples
+without duplicating volatile Notarius contracts.
+
+**Work:**
+
+- Update `docs/config.md` with `pipeline.notarius.references`, its selector-to-
+  source shape, normalization/validation rules, required-by-presence behavior,
+  supported stable source IDs, and campaign/session `spell_catalog_file`
+  precedence and optionality. Keep complete copyable YAML in `examples/`.
+- Update `docs/integrations/notarius.md` to the v0.6.0 baseline and exact
+  repeatable-reference invocation boundary. Explain absolute CLI paths,
+  precedence over configured external paths, the four maintained external D&D
+  slots, the generated-handoff exclusion, and unchanged receipt-v2/ten-lane
+  output compatibility. Link to Notarius's canonical v0.6 CLI and D&D consumer
+  docs instead of copying its target/module matrix.
+- Update operations and troubleshooting with prepared input location,
+  fingerprint/rerun consequences, operator inspection, missing-reference
+  diagnosis, and Notarius undeclared-slot/generated-collision failures. Update
+  internal component documents only with implemented ownership and flow; do not
+  duplicate configuration field definitions there.
+- Add a valid, secret-free sample spell catalog following Notarius v0.6's
+  published overlay schema, add `spell_catalog_file` to the sample campaign, and
+  configure all four external reference bindings in the complete annotated D&D
+  pipeline. Add the same bindings to other Notarius-enabled maintained examples
+  only when their selected pipeline declares them; do not add a Notarius section
+  to examples that intentionally omit extraction.
+- Ensure the maintained command snippets place repeated `--reference` arguments
+  before `--json`, use `party` rather than `roster`, and never show generated
+  handoffs on the CLI. Remove stale v0.5 compatibility wording where it refers to
+  the supported invocation baseline.
+
+**Tests and exit criteria:** Run documentation-link checks and the example
+loader explicitly. Verify every changed example is accepted by strict config
+validation, contains no credentials or private infrastructure values, and has
+one canonical owner for each volatile fact. Search current-behavior docs and
+examples for stale v0.5 invocation wording, deprecated `roster` emission, and
+generated D&D CLI handoff examples. Do not mark the roadmap itself implemented;
+its status remains target-state context until the implementation sprint is
+reviewed and closed.
+
+## Stage 8 — Compatibility And Quality Closure
+
+**Read first:** The feature roadmap, every completed stage diff, the final
+current-behavior docs, `.woodpecker/verify.yml`, `.woodpecker/release.yml`, and
+`.woodpecker/shuffle.yml`. Re-read the tagged Notarius v0.6.0
+`docs/consumers/dnd-pipeline.md`, `docs/cli.md`, and linked spell-catalog overlay
+contract when the sibling checkout is available.
+
+**Depends on:** Stage 7.
+
+**Goal:** Verify the delivered code matches the accepted boundary, remains
+compatible with all ten default D&D artifacts, and is ready for review without
+dead compatibility paths or duplicated policy.
+
+**Work:**
+
+- Audit the final diff against every target-state and out-of-scope statement in
+  the feature roadmap. Confirm only four external prepared sources are exposed,
+  custom selectors remain possible, every configured binding is required, and
+  no Notarius pipeline topology or generated-handoff logic moved into Narratio.
+- Trace initial extract and resume paths to confirm both use the same prepared
+  identity and reference resolution, the adapter is the sole argv builder, and
+  artifact policy is the sole source-to-kind/filename vocabulary. Remove dead
+  helpers, redundant switches, stale fixtures, and low-value duplicate tests
+  found during this review.
+- Confirm the complete D&D example still declares and validates the exact ten
+  output lanes and that analyze can consume those sources after reference-
+  enabled extraction. Confirm empty-reference custom pipelines remain supported.
+- If a local Notarius v0.6.0 binary and its required offline/test configuration
+  are already available, perform a non-credentialed smoke invocation with all
+  four reference flags and record the result in the implementation handoff. Do
+  not download tools, contact paid providers, add a default test dependency, or
+  block completion solely because this supplementary environment is absent.
+- Run the repository baseline plus the scheduled shuffled suite and release
+  cross-build commands:
+
+  ```sh
+  go test ./...
+  go test -race ./...
+  go test -race -shuffle=on -count=3 ./...
+  go vet ./...
+  go build ./...
+  go test ./internal/doccheck
+  go test ./internal/config -run '^TestExamplesLoadAndValidate$'
+  narratio_cross_dir="$(mktemp -d)"
+  CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build -o "$narratio_cross_dir/narratio-linux-amd64" ./cmd/narratio
+  CGO_ENABLED=0 GOOS=darwin GOARCH=amd64 go build -o "$narratio_cross_dir/narratio-darwin-amd64" ./cmd/narratio
+  CGO_ENABLED=0 GOOS=windows GOARCH=amd64 go build -o "$narratio_cross_dir/narratio-windows-amd64.exe" ./cmd/narratio
+  ```
+
+  Cross-builds are compilation evidence only; do not claim native macOS or
+  Windows runtime validation.
+
+**Tests and exit criteria:** Every required command passes, `git diff --check`
+is clean, the worktree contains no unintended generated test artifacts, and the
+implementation is traceably complete against the roadmap. Summarize any
+unavailable supplementary smoke evidence without treating it as a product
+question or silently weakening the default suite.
+
+## Open Questions
+
+None. The feature roadmap and governing decisions above are sufficient to
+implement the plan without additional product or architecture choices.

docs/roadmap/notarius-v0.6-cli-references.md

@@ -0,0 +1,274 @@
+# Notarius v0.6 CLI Reference Integration
+
+## Status
+
+Accepted target state. Delivery sequencing and implementation status are owned
+by [implementation.md](implementation.md).
+
+## Purpose
+
+Upgrade Narratio's extraction boundary to the Notarius v0.6.0 subprocess
+contract and supply session reference documents explicitly with repeatable
+`--reference selector=path` arguments.
+
+The maintained D&D integration must make the prepared party roster, player
+context, glossary, and optional spell catalog available to every compatible
+Notarius target. Notarius must continue to own pipeline topology, reference-slot
+compatibility, generated artifact handoffs, prompts, and D&D schemas. Narratio
+owns selection and preparation of its external reference files, exact CLI
+invocation, provenance, and extraction reuse correctness.
+
+## Current State And Gap
+
+Narratio currently invokes Notarius as:
+
+```text
+notarius run <pipeline_id> --config <config_path> --input <transcript> --output-dir <staging_dir> --json
+```
+
+The `prepare` stage already materializes campaign/session party, players, and
+glossary files under the session `inputs/` directory, but `extract` does not
+pass them to Notarius. Narratio also has no stable spell-catalog input. As a
+result, a Notarius deployment must duplicate these paths in its own
+configuration, cannot reliably receive session overrides, and may extract
+without the same campaign context supplied to Narratio's analysis stage.
+
+Notarius v0.6.0 makes an unqualified CLI selector pipeline-scoped. For example,
+`--reference party=/absolute/path/party.yml` supplies the file to every
+selected target that declares `party`. Scoped selectors remain available for
+exceptional overrides. CLI paths are resolved from the Notarius process working
+directory, so subprocess callers are expected to provide absolute paths.
+
+The v0.6.0 receipt, index, warning, diagnostic, and ten-lane D&D artifact
+contracts remain compatible with Narratio's current v0.5 integration. This
+feature changes the invocation and input-provenance contract rather than the
+accepted output inventory.
+
+## User Outcome
+
+With the maintained complete D&D configuration, an operator can declare the
+campaign reference sources once in Narratio. For each extraction Narratio will:
+
+1. materialize the effective campaign/session files during `prepare`;
+2. resolve those prepared files by stable Narratio source ID;
+3. pass absolute paths for `party`, `players`, `glossary`, and, when configured,
+   `spell_catalog` to Notarius through repeatable CLI arguments;
+4. fail before launching Notarius when a configured reference is unavailable;
+5. rerun extraction when a selector, source binding, or reference file changes;
+   and
+6. retain bounded reference identities and checksums for diagnosis and
+   provenance without copying reference contents into manifest metadata.
+
+Session-level stable-input overrides must flow through the same mechanism. A
+custom Notarius pipeline may bind different external slots without requiring a
+Narratio code change.
+
+## Chosen Architecture
+
+### Explicit Reference Bindings
+
+Extend `pipeline.notarius` with an explicit map from a Notarius CLI selector to
+a prepared Narratio input source:
+
+```yaml
+notarius:
+  enabled: true
+  binary: notarius
+  config_path: /usr/local/etc/notarius/config.yml
+  pipeline_id: dnd-session
+  working_directory: /usr/local/etc/notarius
+  references:
+    party: narratio.input.party
+    players: narratio.input.players
+    glossary: narratio.input.glossary
+    spell_catalog: narratio.input.spell_catalog
+  outputs:
+    # Existing required lane contracts remain unchanged.
+```
+
+Each configured binding is required. An operator who does not maintain an
+optional Notarius reference, such as a spell catalog, omits that binding. This
+keeps missing-input behavior explicit and avoids a second required/optional
+policy inside each entry.
+
+The maintained complete D&D example will show all four external reference
+slots. The three existing campaign context bindings use the canonical `party`,
+`players`, and `glossary` spellings. Narratio will not emit the deprecated
+`roster` alias.
+
+The binding is deliberately source-based rather than path-based. Pipeline
+configuration should not reconstruct session workspace paths or bypass
+`prepare`; it names the stable input whose effective campaign/session value is
+already owned by Narratio. The map also avoids hard-coded behavior keyed to the
+literal `dnd-session` pipeline ID, preserving custom-pipeline support.
+
+Narratio accepts the selector forms published by Notarius v0.6.0:
+
+- `slot`;
+- `chunk.slot`;
+- `lane.slot`; and
+- `lane.extract.slot`, `lane.merge.slot`, or `lane.normalize.slot`.
+
+Configuration validation will reject empty or structurally invalid selectors,
+selectors containing `=`, unsupported source IDs, and duplicate YAML keys.
+Notarius remains authoritative for whether a selected target actually declares
+the slot and whether a file satisfies that slot's media type and size contract.
+Narratio will not duplicate the Notarius module registry.
+
+### Stable Reference Inputs
+
+Continue to use the existing prepared sources and canonical files:
+
+| Narratio source | Prepared file | Notarius slot |
+| --- | --- | --- |
+| `narratio.input.party` | `inputs/party.yml` | `party` |
+| `narratio.input.players` | `inputs/players.yml` | `players` |
+| `narratio.input.glossary` | `inputs/glossary.yml` | `glossary` |
+| `narratio.input.spell_catalog` | `inputs/spell_catalog.json` | `spell_catalog` |
+
+Add optional `spell_catalog_file` fields to campaign and session inputs, with
+the existing campaign-default/session-override resolution behavior. When
+provided, `prepare` copies it into the session input area and records its
+origin and checksum consistently with the other stable inputs. The prepared
+filename remains JSON so Notarius can apply its published spell-catalog media
+contract.
+
+The new source must be added everywhere stable inputs are enumerated: strict
+configuration decoding and merging, validation, prepare materialization,
+artifact policy/source descriptions, operator inspection, manifest input
+records, examples, and canonical documentation. It remains optional at the
+campaign level; a configured Notarius binding makes it mandatory for that
+extraction.
+
+Extract and analyze should use one shared prepared-input source resolver rather
+than maintain separate source-to-filename tables. The resolver must return an
+absolute, regular, non-empty file beneath the current session workspace and
+produce actionable `prepare --force` guidance when a configured source is
+missing. It must not fall back to the original campaign path after preparation.
+
+### Adapter Request And CLI Construction
+
+Extend the transport-neutral Notarius run request with an ordered collection of
+resolved reference bindings. Each binding contains only its selector and
+absolute prepared-file path. The extraction stage resolves source IDs and file
+identity; the subprocess adapter validates and serializes the request.
+
+The production command becomes:
+
+```text
+notarius run <pipeline_id>
+  --config <config_path>
+  --input <trimmed_json>
+  --output-dir <staging_dir>
+  --reference party=<absolute_prepared_party_path>
+  --reference players=<absolute_prepared_players_path>
+  --reference glossary=<absolute_prepared_glossary_path>
+  --reference spell_catalog=<absolute_prepared_spell_catalog_path>
+  --json
+```
+
+Only configured bindings are emitted. Selectors are sorted before request
+construction so argument order, tests, logs, and fingerprints are deterministic.
+Arguments are passed directly to the subprocess without shell interpretation;
+paths containing spaces or platform-specific separators remain one argument.
+
+CLI bindings intentionally override matching external paths in the deployed
+Notarius configuration. Narratio must not pass `--without-reference` and must
+not synthesize CLI bindings for `location_registry`, `item_registry`,
+`npc_registry`, `scene_descriptions`, `combat_turns`, or `npc_occurrences`.
+Those are generated same-run artifact handoffs in the complete D&D pipeline and
+remain entirely under Notarius configuration and execution control. A custom
+configuration that collides an external CLI binding with a generated handoff is
+expected to fail with Notarius's normal resolution error.
+
+### Fingerprints, Resume, And Provenance
+
+Reference identity is part of the extraction input contract. The extraction
+fingerprint and resume validator must include, in deterministic selector order:
+
+- the selector;
+- the configured Narratio source ID;
+- the resolved prepared path identity; and
+- the prepared file's content checksum and size.
+
+This is required even though Notarius generates a prompt session ID from the
+input module and transcript bytes: Notarius intentionally does not include
+references in that identifier. Narratio must therefore prevent an old
+extraction from being reused after a roster, player list, glossary, spell
+catalog, selector, or source mapping changes.
+
+A changed reference makes the prior `extract` result non-reusable and follows
+Narratio's normal downstream invalidation rules. A failed reference-resolution
+or checksum check also prevents reuse; it must not silently accept the prior
+bundle.
+
+Successful extract metadata should record a bounded, deterministic list of
+selector, source ID, workspace-relative path, checksum, and size. It must not
+record reference contents, original absolute operator paths, or values from the
+files. Existing receipt and bundle provenance behavior remains unchanged.
+
+### Error And Compatibility Behavior
+
+Narratio's documented minimum supported Notarius version becomes v0.6.0 for an
+enabled reference binding. Compatibility remains contract-based rather than
+dependent on parsing `notarius --version`: an older or incompatible executable
+will fail at the CLI boundary with captured diagnostics.
+
+Errors must identify the responsible selector and Narratio source without
+including file contents. Configuration errors are reported before pipeline
+execution. Missing, empty, non-regular, unsafe, or unreadable prepared files
+fail extraction before the Notarius subprocess starts. Notarius continues to
+report undeclared slots, media incompatibility, size limits, required-slot
+failures, and generated-handoff collisions.
+
+When Notarius is disabled, extraction retains its current explicit skip
+behavior and does not resolve reference inputs. Receipt v2 ingestion, bundle
+confinement, ten-lane selection, and downstream artifact source IDs are not
+otherwise changed by this feature.
+
+## Target End State
+
+Narratio and Notarius have a clear orchestration boundary:
+
+- `prepare` owns the effective, immutable session copies of external campaign
+  context;
+- `extract` maps configured stable source IDs to Notarius v0.6 CLI selectors,
+  supplies absolute file paths, and owns reuse/provenance policy;
+- the Notarius adapter owns exact subprocess serialization and supported result
+  decoding;
+- Notarius owns slot compatibility, reference precedence within its pipeline,
+  generated artifact handoffs, and output schemas; and
+- `analyze` consumes the resulting ten structured lane artifacts exactly as it
+  does today.
+
+The maintained complete D&D workflow passes party, players, glossary, and spell
+catalog context from the same prepared session inputs used elsewhere in
+Narratio. Updating any of those documents deterministically causes fresh
+extraction, and operators can diagnose the effective bindings without exposing
+file contents.
+
+## Out Of Scope
+
+- Reproducing Notarius pipeline, lane, binding, or media-type validation in
+  Narratio.
+- Passing or overriding Notarius generated artifact handoffs.
+- Adding `--without-reference`, Notarius resume/recompute controls, lane
+  selection, model selection, profile selection, or session-ID overrides.
+- Changing the ten accepted D&D lane contracts or the Scriptorium analysis
+  design.
+- Reading reference payloads into Narratio manifests or logs.
+- Automatically running `notarius config validate` for every session.
+
+## Settled Policy Choices
+
+The implementation must preserve these choices unless implementation evidence
+shows a contract conflict:
+
+- explicit selector-to-source mappings are preferred over pipeline-ID-specific
+  defaults;
+- every configured mapping is required;
+- `spell_catalog_file` is optional until a mapping requests its prepared
+  source;
+- the complete D&D example demonstrates all four external references; and
+- Notarius v0.6.0 is the minimum supported CLI contract for reference-enabled
+  extraction.

docs/troubleshooting.md

@@ -0,0 +1,543 @@
+# Troubleshooting
+
+Operational diagnosis guide for common Narratio failures.
+
+## Config file not found
+
+Symptom:
+
+- command fails to resolve `pipeline.yml`, `campaign.yml`, or `session.yml`.
+
+Likely causes:
+
+- missing files in default search paths;
+- wrong campaign selection;
+- omitted explicit flags.
+
+Diagnostics:
+
+```bash
+narratio session plan 2026-04-04
+```
+
+Safe fix:
+
+- pass explicit `--config`, `--campaign` or `--campaign-file`, and `--session`.
+
+Relevant reference: [Configuration discovery](./config.md#discovery-and-selection).
+
+## Session template placeholders rejected
+
+Symptom:
+
+- load error says session file must be concrete or contains `{{ ... }}` placeholders.
+
+Likely cause:
+
+- using template content as runtime session config.
+
+Diagnostics:
+
+```bash
+narratio session validate 2026-04-04 --session /path/session.yml
+```
+
+Safe fix:
+
+- generate concrete session YAML with `narratio session init`.
+
+Relevant reference: [Operations: Session Initialization](./operations.md#session-initialization).
+
+## Strict decode or schema validation failure
+
+Symptom:
+
+- unknown field / invalid value error during config load.
+
+Likely cause:
+
+- stale field name, typo, invalid enum, or invalid duration/path format.
+
+Diagnostics:
+
+```bash
+narratio session plan 2026-04-04 --config /path/pipeline.yml --campaign-file /path/campaign.yml --session /path/session.yml
+```
+
+Safe fix:
+
+- align config with [Configuration](./config.md) and the
+  [maintained examples](../examples/README.md).
+
+Relevant reference: [Configuration](./config.md).
+
+## Audio mode conflict
+
+Symptom:
+
+- validation fails on session audio configuration.
+
+Likely cause:
+
+- configured both local and S3 session audio inputs.
+
+Diagnostics:
+
+```bash
+narratio session validate 2026-04-04
+```
+
+Safe fix:
+
+- use local mode (`audio_dir` or `audio_files`) or S3 mode (`audio_s3.prefix`), not both.
+
+Relevant reference: [Session configuration](./config.md#session).
+
+## `--artifacts` selection error
+
+Symptom:
+
+- unknown artifact key or invalid `--artifacts` usage.
+
+Likely causes:
+
+- key not defined in `pipeline.scriptorium.artifacts`;
+- empty list entry (for example trailing comma);
+- `run-stage` used with non-`analyze`/`publish` target.
+
+Diagnostics:
+
+```bash
+narratio session artifacts 2026-04-04
+```
+
+Safe fix:
+
+- provide only configured keys and use `--artifacts` with supported commands/stages.
+
+Relevant reference: [CLI artifact selection](./cli.md).
+
+## Notarius executable missing
+
+Symptom:
+
+- extraction fails while resolving or starting the Notarius executable.
+
+Likely causes:
+
+- `pipeline.notarius.binary` is not installed, executable, or on `PATH`;
+- a configured executable path is wrong.
+
+Safe fix:
+
+- install a compatible Notarius release or correct the binary setting, then
+  rerun extraction.
+
+Relevant references: [Notarius configuration](./config.md#notarius-output-entries)
+and [Notarius integration](./integrations/notarius.md).
+
+## Notarius exits nonzero
+
+Symptom:
+
+- extraction reports a Notarius exit error instead of a receipt.
+
+Diagnostics:
+
+- inspect `runs/{run_id}/extract/notarius.stderr.log`; stdout is reserved for
+  the receipt and is not merged with diagnostics.
+
+Safe fix:
+
+- correct the reported Notarius pipeline, input, provider, or configuration
+  failure and rerun extraction. Do not edit a staged output bundle into place.
+
+After a failed replacement, an older immutable bundle may still exist even
+though the current session manifest has no successful extraction payload. This
+is expected audit state, not a signal to relink the old bundle manually.
+
+Relevant reference: [Operations: Extraction Workflow](./operations.md#extraction-workflow).
+
+## Prepared Notarius reference missing or inconsistent
+
+Symptom:
+
+- extraction or resume validation reports that a configured reference source is
+  unavailable, unsafe, empty, or checksum-inconsistent and recommends
+  `prepare --force`.
+
+Likely causes:
+
+- `prepare` has not run since the campaign/session stable input changed;
+- the configured source file is missing;
+- a prepared `inputs/` file or its manifest record was modified independently;
+- a spell-catalog binding exists without an effective `spell_catalog_file`.
+
+Diagnostics:
+
+```bash
+narratio session status 2026-04-04
+narratio session validate 2026-04-04
+```
+
+Safe fix:
+
+- correct the campaign/session input path, then refresh canonical prepared
+  evidence before extraction:
+
+```bash
+narratio run-stage prepare 2026-04-04 --force
+```
+
+Do not point Notarius directly at the original source path or edit the manifest
+checksum. Relevant references: [Notarius reference configuration](./config.md#notarius-reference-bindings)
+and [Operations: Extraction Workflow](./operations.md#extraction-workflow).
+
+## Notarius reference selector or generated-handoff collision
+
+Symptom:
+
+- Notarius exits nonzero with an undeclared reference-slot, incompatible media,
+  or external/generated reference collision error.
+
+Likely causes:
+
+- a selector does not identify a slot declared by the selected Notarius target;
+- a prepared file does not satisfy that slot's Notarius media contract; or
+- a CLI binding attempts to replace a same-run generated D&D handoff.
+
+Safe fix:
+
+- compare external bindings with the selected Notarius pipeline's canonical
+  consumer documentation;
+- keep only campaign-owned external slots on the CLI; and
+- leave registry, scene, combat, and occurrence handoffs to Notarius pipeline
+  composition.
+
+Narratio validates selector structure and prepared evidence, while Notarius
+owns slot declarations, media compatibility, and generated-handoff conflicts.
+Relevant reference: [Notarius integration](./integrations/notarius.md).
+
+## Atomic Notarius promotion unsupported
+
+Symptom:
+
+- extraction fails with `atomic no-replace directory promotion is unsupported`
+  before a durable bundle or temporary promotion tree is created.
+
+Likely cause:
+
+- Narratio is running on an operating system other than Linux, macOS, or
+  Windows, where the required atomic no-replace directory primitive has not
+  been implemented and verified.
+
+Safe fix:
+
+- run extraction on Linux, macOS, or Windows. Do not replace the atomic commit
+  with a manual copy or move; the session manifest must never observe a partial
+  or overwritten bundle.
+
+This is an extraction-specific platform boundary, not a support statement for
+unrelated Narratio workflows. See
+[Operations: Extraction Workflow](./operations.md#extraction-workflow).
+
+## Notarius receipt or index incompatible
+
+Symptom:
+
+- extraction rejects the receipt schema, pipeline identity, bundle/index path,
+  lane descriptor, or payload path even though Notarius exited successfully.
+
+Likely causes:
+
+- Narratio and Notarius versions disagree on their consumer contract;
+- the configured pipeline or lane constraints are stale;
+- output paths escape the bundle or traverse symlinks.
+
+Safe fix:
+
+- compare installed Notarius output with the canonical Notarius contracts,
+  including receipt `index_file: index.json` and index management names
+  `manifest.json`, `rejected.json`, `warnings.json`, and `diagnostics.json`; align
+  `pipeline.notarius` constraints and rerun. Do not bypass confinement or schema
+  checks.
+
+Relevant reference: [Notarius integration](./integrations/notarius.md).
+
+## Required Notarius lane rejected or missing
+
+Symptom:
+
+- extraction fails because a configured lane is rejected, missing, duplicated,
+  or incompatible, including after a zero exit.
+
+Safe fix:
+
+- inspect the Notarius diagnostic log and bundle rejection/warning information;
+- correct the Notarius module or the exact declared lane contract;
+- remove an output declaration only if downstream consumers genuinely no longer
+  require that source, then rerun extraction.
+
+Every configured output is required. Narratio does not promote a partial result.
+
+## Extraction resume invalidated
+
+Symptom:
+
+- a previously successful extraction runs again during ordinary continuation.
+
+Likely causes:
+
+- the executable/config path, pipeline ID, timeout, working directory, or
+  configured output contracts changed;
+- a configured prepared reference selector, source, path, checksum, or byte
+  size changed;
+- the durable bundle, index, lane set, provenance, regular-file status, or
+  checksum no longer validates.
+
+Safe fix:
+
+- allow the automatic rerun after verifying the current configuration. Treat
+  an unsafe path or symlink error as filesystem corruption or tampering and
+  investigate it rather than replacing files manually.
+
+## Notarius transitive configuration changed
+
+Symptom:
+
+- Notarius profiles, prompts, modules, imported files, or references changed,
+  but Narratio still considers the previous extraction resumable.
+
+Safe fix:
+
+```bash
+narratio run-stage extract 2026-04-04 --force
+```
+
+Narratio fingerprints its invocation contract and prepared Narratio reference
+identities, not the contents of other transitive Notarius inputs. Always force
+extraction after changing those external inputs; downstream
+successful stages are then marked stale normally.
+
+Relevant reference: [Operations: Extraction Workflow](./operations.md#extraction-workflow).
+
+## Previous-session artifact input missing
+
+Symptom:
+
+- prepare/analyze fails due to missing required previous-session artifact cache input.
+
+Likely causes:
+
+- missing `session.previous_session_id`;
+- previous artifact not restored/published for source session.
+
+Diagnostics:
+
+```bash
+narratio session validate 2026-04-04
+narratio session status 2026-04-04
+```
+
+Safe fix:
+
+```bash
+narratio session restore 2026-04-04
+```
+
+or rerun prepare after correcting session config:
+
+```bash
+narratio run-stage prepare 2026-04-04 --force
+```
+
+Relevant reference: [Operations: Restore Workflow](./operations.md#restore-workflow).
+
+## Session lock conflict (`.lock`)
+
+Symptom:
+
+- command fails acquiring session lock.
+
+Likely causes:
+
+- another process is running for the same session;
+- a process still holds the operating-system lock while it is shutting down.
+
+Diagnostics:
+
+```bash
+ls -l {workspace.root}/work/{campaign}/{session_id}/.lock
+ps aux | grep narratio
+```
+
+Safe fix:
+
+- wait for active process completion;
+- retry after an interrupted holder has exited; the kernel releases its lock
+  even though the `.lock` metadata file remains for inspection.
+
+Relevant reference: [Operations: Local State Layout](./operations.md#local-state-layout).
+
+## Restore conflict without `--force`
+
+Symptom:
+
+- restore fails with conflict count.
+
+Likely cause:
+
+- local durable files differ from remote restore sources.
+
+Diagnostics:
+
+```bash
+narratio session restore 2026-04-04 --dry-run
+```
+
+Safe fix:
+
+- review conflicts;
+- rerun with `--force` only when remote state should overwrite local.
+
+Relevant reference: [Operations: Restore Workflow](./operations.md#restore-workflow).
+
+## Restore current-state discovery failure
+
+Symptom:
+
+- restore cannot find current pointer or current manifest.
+
+Likely causes:
+
+- no committed publish current state;
+- storage credentials or connectivity failure.
+
+Diagnostics:
+
+```bash
+narratio session status 2026-04-04
+narratio session restore 2026-04-04 --dry-run
+```
+
+Safe fix:
+
+- resolve storage/auth issue;
+- republish from healthy local state if current pointer is missing.
+
+Relevant reference: [Operations: Publish Workflow](./operations.md#publish-workflow).
+
+## Publish output failure
+
+Symptom:
+
+- publish fails on missing required source, upload error, or commit write.
+
+Likely causes:
+
+- required source file not produced;
+- lock/state expectations mismatch;
+- remote storage failure.
+
+Diagnostics:
+
+```bash
+narratio session artifacts 2026-04-04 --remote
+narratio session status 2026-04-04
+narratio run-stage publish 2026-04-04 --force
+```
+
+Safe fix:
+
+- regenerate missing sources by rerunning prerequisite stages;
+- correct publish source/destination rules;
+- retry after storage failure is resolved.
+
+Relevant reference: [Publish configuration](./config.md#publish-configuration-summary).
+
+## Render markdown source missing
+
+Symptom:
+
+- analyze or publish fails because `narratio.transcript.final_markdown` or `narratio.transcript.final_trimmed_markdown` is unavailable.
+
+Likely causes:
+
+- render stage was not executed after transcript changes;
+- render stage failed before producing canonical markdown outputs.
+
+Diagnostics:
+
+```bash
+narratio session status 2026-04-04
+```
+
+Safe fix:
+
+- rerun render and then retry downstream stage(s):
+
+```bash
+narratio run-stage render 2026-04-04 --force
+narratio run-stage analyze 2026-04-04 --force
+```
+
+Relevant reference: [Operations: Stage Execution](./operations.md#stage-execution-and-continuation-behavior).
+
+## Secrets or storage credential failure
+
+Symptom:
+
+- object-store command fails at initialization/auth.
+
+Likely causes:
+
+- invalid `pipeline.secrets.env_dir`;
+- missing credential environment variables;
+- invalid S3 endpoint/bucket settings.
+
+Diagnostics:
+
+```bash
+ls -la /path/to/secrets_dir
+env | sed 's/=.*//' | grep -E 'OBJECT_STORAGE|AWS|AUDITA|SCRIPTORIUM'
+```
+
+Safe fix:
+
+- correct secret-file path and permissions;
+- provide required env vars;
+- keep secret values out of YAML.
+
+Relevant reference: [Secrets](./config.md#secrets-handling).
+
+## S3 audio prepare failure
+
+Symptom:
+
+- prepare fails listing/downloading session S3 audio.
+
+Likely causes:
+
+- incorrect `session.inputs.audio_s3.prefix`;
+- no matching `.flac` objects;
+- storage connectivity or permissions failure.
+
+Diagnostics:
+
+```bash
+narratio session validate 2026-04-04
+```
+
+Safe fix:
+
+- verify prefix contents and storage access;
+- keep session audio mode consistent.
+
+Relevant reference: [Operations](./operations.md).
+
+## References
+
+- [docs/cli.md](./cli.md)
+- [docs/config.md](./config.md)
+- [docs/operations.md](./operations.md)
+- [docs/internal/stage-publish.md](./internal/stage-publish.md)

examples/README.md

@@ -0,0 +1,50 @@
+# Maintained Examples
+
+These files are safe, copyable starting points for Narratio configuration and
+input structure. Replace placeholder identifiers, storage names, integration
+URLs, and paths for the target environment. Field meanings and defaults belong
+in the [configuration reference](../docs/config.md).
+
+## Pipeline Configuration
+
+- [Minimal pipeline](pipeline.minimal.yml): campaign discovery plus the required
+  WhisperX URL.
+- [Production-shaped pipeline](pipeline.production.yml): S3 storage, publish,
+  external tools, and configured Scriptorium artifacts.
+- [Full annotated pipeline](pipeline.full.annotated.yml): every implemented
+  pipeline section with explanatory comments.
+- [Extraction subset pipeline](pipeline.extraction-subset.yml): a focused
+  Scriptorium artifact consuming only three declared Notarius lanes.
+
+The existing `internal/config` example test loads and validates each pipeline
+with the sample campaign and a compatible local- or S3-audio session.
+
+## Campaign And Session Configuration
+
+- [Sample campaign](campaigns/sample-campaign/campaign.yml), its
+  [session template](campaigns/sample-campaign/session.template.yml), and its
+  adjacent stable inputs provide a complete campaign directory shape.
+- [Local-audio session](session.local-audio.yml) and
+  [S3-audio session](session.s3-audio.yml) are concrete session files.
+- [Session template](session.template.yml) and the campaign-local equivalent
+  demonstrate the narrow placeholder syntax consumed by `session init`; they
+  are templates, not runtime session files.
+
+## Input Fixtures
+
+- [Speakers](speakers.yml), [autocorrect](autocorrect.yml), and
+  [glossary](glossary.yml) show the standalone input shapes.
+- The sample campaign references its local
+  [speakers](campaigns/sample-campaign/speakers.yml),
+  [autocorrect](campaigns/sample-campaign/autocorrect.yml),
+  [glossary](campaigns/sample-campaign/glossary.yml),
+  [players](campaigns/sample-campaign/players.yml), and
+  [party](campaigns/sample-campaign/party.yml) fixtures, plus an optional
+  [spell-catalog overlay](campaigns/sample-campaign/spell_catalog.json) that
+  follows the Notarius v0.6 contract.
+- [Sample speaker audio](audio/sample-speaker.flac) is a text placeholder that
+  reserves the expected filename and directory shape. Replace it with a real
+  FLAC file before running transcription.
+
+The examples contain environment-variable names but no credential values. They
+use fictional campaign content and reserved example domains.

examples/campaigns/sample-campaign/autocorrect.yml

@@ -0,0 +1 @@
+[]

examples/campaigns/sample-campaign/campaign.yml

@@ -0,0 +1,9 @@
+campaign_id: sample-campaign
+session_template_file: ./session.template.yml
+inputs:
+  speakers_file: ./speakers.yml
+  autocorrect_file: ./autocorrect.yml
+  glossary_file: ./glossary.yml
+  players_file: ./players.yml
+  party_file: ./party.yml
+  spell_catalog_file: ./spell_catalog.json

examples/campaigns/sample-campaign/glossary.yml

@@ -0,0 +1 @@
+[]

examples/campaigns/sample-campaign/party.yml

@@ -0,0 +1,2 @@
+- name: Example Hero
+  type: pc

examples/campaigns/sample-campaign/players.yml

@@ -0,0 +1,2 @@
+- name: Example Player
+  role: player

examples/campaigns/sample-campaign/session.template.yml

@@ -0,0 +1,3 @@
+session_id: "{{ session_id }}"
+inputs:
+  audio_dir: ./audio

examples/campaigns/sample-campaign/speakers.yml

@@ -0,0 +1,5 @@
+match:
+  - speaker: "Example Speaker"
+    match:
+      - "Example_Speaker"
+      - "Example"

examples/campaigns/sample-campaign/spell_catalog.json

@@ -0,0 +1,18 @@
+{
+  "schema_version": "notarius.dnd.spell-catalog-overlay.v1",
+  "catalogs": [
+    {
+      "id": "narratio.sample-campaign",
+      "ruleset": "dnd-5e-2014",
+      "source": {
+        "title": "Narratio sample campaign spell names"
+      },
+      "spells": [
+        {
+          "name": "Aegis of Emberfall",
+          "aliases": ["Emberfall Aegis"]
+        }
+      ]
+    }
+  ]
+}

examples/pipeline.extraction-subset.yml

@@ -0,0 +1,59 @@
+# Purpose-specific extraction example: a Scriptorium session brief consumes
+# only the three Notarius lanes it needs.
+
+campaigns:
+  root: /usr/local/share/narratio/campaigns
+  default_campaign_id: sample-campaign
+
+whisperx:
+  transcribe_url: "https://transcription.example.com/transcribe"
+
+notarius:
+  enabled: true
+  binary: notarius
+  config_path: /usr/local/etc/notarius/config.yml
+  pipeline_id: dnd-session
+  timeout: 3h
+  references:
+    glossary: narratio.input.glossary
+    party: narratio.input.party
+    players: narratio.input.players
+    spell_catalog: narratio.input.spell_catalog
+  outputs:
+    npc_registry:
+      lane_id: npc-registry
+      media_type: application/json
+      schema_id: notarius.dnd.npc_registry
+      schema_version: v1
+      module_key: dnd/npc-registry
+    location_registry:
+      lane_id: location-registry
+      media_type: application/json
+      schema_id: notarius.dnd.location_registry
+      schema_version: v1
+      module_key: dnd/location-registry
+    scene_descriptions:
+      lane_id: scene-descriptions
+      media_type: application/json
+      schema_id: notarius.dnd.scene_descriptions
+      schema_version: v1
+      module_key: dnd/scene-descriptions
+
+scriptorium:
+  binary: scriptorium
+  config_path: /usr/local/etc/scriptorium/config.yml
+  artifacts:
+    session_brief:
+      enabled: true
+      prompt_id: dnd.session_brief
+      output_path: artifacts/session_brief.md
+      inputs:
+        npcs:
+          source: narratio.extraction.npc_registry
+          required: true
+        locations:
+          source: narratio.extraction.location_registry
+          required: true
+        scenes:
+          source: narratio.extraction.scene_descriptions
+          required: true

examples/pipeline.full.annotated.yml

@@ -0,0 +1,271 @@
+# Full annotated pipeline example for implemented Narratio config fields.
+# Values are safe placeholders and must be adapted per environment.
+
+workspace:
+  # Optional: defaults to /var/lib/narratio.
+  root: /var/lib/narratio/workspace
+  # Optional: remove run-scoped workdir after successful publish commit.
+  cleanup_after_publish: false
+
+# Optional: local secret file loader (directory of ENV_VAR_NAME files).
+# secrets:
+#   env_dir: ./secrets
+
+storage:
+  # Defaults to "local". Use "s3" explicitly for publish + S3 audio workflows.
+  backend: s3
+  s3:
+    # Required when using S3 audio or S3 publish uploads.
+    bucket: my-dnd-archive
+    # Optional; defaults to "dnd".
+    root_prefix: dnd
+    # Optional region/endpoint settings.
+    region: us-east-1
+    endpoint: ""
+    force_path_style: false
+    # Optional; defaults shown explicitly.
+    access_key_id_env: OBJECT_STORAGE_KEY_ID
+    secret_access_key_env: OBJECT_STORAGE_KEY
+
+campaigns:
+  # Optional; defaults to /usr/local/share/narratio/campaigns.
+  root: /usr/local/share/narratio/campaigns
+  # Optional command default when --campaign is omitted.
+  default_campaign_id: sample-campaign
+
+spool:
+  # Optional; defaults to /var/spool/narratio.
+  root: /var/spool/narratio
+  # Optional cleanup of run-scoped spool audio after successful publish commit.
+  delete_audio_after_publish: false
+
+publish:
+  # Optional booleans; defaults are true.
+  enabled: true
+  upload_run: true
+  # Optional publish output rules; sources use Narratio artifact source IDs.
+  outputs:
+    - source: narratio.transcript.final_trimmed
+      dest: transcripts/final.trimmed.json
+      required: true
+    - source: narratio.transcript.final_markdown
+      dest: transcripts/final.md
+      required: true
+    - source: narratio.transcript.final_trimmed_markdown
+      dest: transcripts/final.trimmed.md
+      required: true
+    - source: narratio.artifact.session_recap
+      dest: artifacts/session_recap.md
+      required: true
+    - source: narratio.artifact.player_handout
+      dest: artifacts/player_handout.md
+      required: false
+    # Extraction lanes publish only when named explicitly; the bundle and index
+    # are never implicit publish sources.
+    - source: narratio.extraction.npc_registry
+      dest: artifacts/extraction/npc-registry.json
+      required: true
+
+whisperx:
+  # Required.
+  transcribe_url: "https://transcription.example.com/transcribe"
+  # Optional overrides; defaults shown explicitly.
+  language: en
+  timeout: 30m
+  retries: 3
+  retry_delay: 2s
+  concurrency: 2
+
+seriatim:
+  # Optional overrides; defaults shown explicitly.
+  binary: seriatim
+  timeout: 10m
+  output_schema: seriatim-intermediate
+  coalesce_gap: 3.0
+  report: true
+  env:
+    # Optional advanced tuning; set only when needed.
+    overlap_word_run_gap: 1.0
+    overlap_word_run_reorder_window: 1.0
+    backchannel_max_duration: 2.0
+    filler_max_duration: 1.25
+
+audita:
+  # Optional overrides; defaults shown explicitly where applicable.
+  binary: audita
+  timeout: 3h
+  llm_api_key_env: AUDITA_LLM_API_KEY
+  modules: [glossary, homophones, spoken_word, grammar]
+  base_url: ""
+  model: ""
+  total_llm_concurrency: 2
+  proposal_llm_concurrency: 1
+  validation_model: ""
+  validation_llm_concurrency: 1
+  transcript_description: ""
+  config_path: /usr/local/etc/audita/config.yml
+  output_schema: audita-v1
+  work_dir_retention: auto
+  report: true
+
+normalize:
+  # Optional; defaults shown explicitly.
+  output_path: transcripts/final.json
+  output_schema: seriatim-intermediate
+  report: true
+
+trim:
+  # Optional; defaults shown explicitly.
+  enabled: true
+  output_path: transcripts/final.trimmed.json
+  bounds:
+    prompt_id: dnd.session_bounds
+    profile_id: ""
+    transcript_input_name: transcript
+    output_path: artifacts/session_bounds.json
+    timeout: 10m
+    render_debug: false
+  seriatim:
+    report: false
+
+notarius:
+  # Optional structured extraction between trim and render.
+  enabled: true
+  binary: notarius
+  config_path: /usr/local/etc/notarius/config.yml
+  pipeline_id: dnd-session
+  timeout: 3h
+  working_directory: /usr/local/etc/notarius
+  # External campaign references use prepared Narratio source IDs. Omit an
+  # optional binding when the selected Notarius pipeline does not need it.
+  references:
+    glossary: narratio.input.glossary
+    party: narratio.input.party
+    players: narratio.input.players
+    spell_catalog: narratio.input.spell_catalog
+  # Each key creates source narratio.extraction.<key>. These constraints match
+  # the current Notarius D&D lane contracts; update them with Notarius.
+  outputs:
+    item_registry:
+      lane_id: item-registry
+      media_type: application/json
+      schema_id: notarius.dnd.item_registry
+      schema_version: v1
+      module_key: dnd/item-registry
+    npc_registry:
+      lane_id: npc-registry
+      media_type: application/json
+      schema_id: notarius.dnd.npc_registry
+      schema_version: v1
+      module_key: dnd/npc-registry
+    location_registry:
+      lane_id: location-registry
+      media_type: application/json
+      schema_id: notarius.dnd.location_registry
+      schema_version: v1
+      module_key: dnd/location-registry
+    scene_descriptions:
+      lane_id: scene-descriptions
+      media_type: application/json
+      schema_id: notarius.dnd.scene_descriptions
+      schema_version: v1
+      module_key: dnd/scene-descriptions
+    item_occurrences:
+      lane_id: item-occurrences
+      media_type: application/json
+      schema_id: notarius.dnd.item_occurrences
+      schema_version: v1
+      module_key: dnd/item-occurrences
+    spells:
+      lane_id: spells
+      media_type: application/json
+      schema_id: notarius.dnd.spells
+      schema_version: v1
+      module_key: dnd/spells
+    combat_turns:
+      lane_id: combat-turns
+      media_type: application/json
+      schema_id: notarius.dnd.combat_turns
+      schema_version: v1
+      module_key: dnd/combat-turns
+    npc_occurrences:
+      lane_id: npc-occurrences
+      media_type: application/json
+      schema_id: notarius.dnd.npc_occurrences
+      schema_version: v1
+      module_key: dnd/npc-occurrences
+    location_occurrences:
+      lane_id: location-occurrences
+      media_type: application/json
+      schema_id: notarius.dnd.location_occurrences
+      schema_version: v1
+      module_key: dnd/location-occurrences
+    enemy_events:
+      lane_id: enemy-events
+      media_type: application/json
+      schema_id: notarius.dnd.enemy_events
+      schema_version: v1
+      module_key: dnd/enemy-events
+
+scriptorium:
+  binary: scriptorium
+  config_path: /usr/local/etc/scriptorium/config.yml
+  timeout: 10m
+  render_debug: false
+  artifacts:
+    # Configured artifact keys map to source IDs narratio.artifact.<key>.
+    session_recap:
+      enabled: true
+      prompt_id: dnd.session_recap
+      profile_id: local-fast
+      output_path: artifacts/session_recap.md
+      timeout: 10m
+      inputs:
+        transcript:
+          source: narratio.transcript.final_trimmed
+          required: true
+        previous_recap:
+          source: narratio.previous_session.artifact.session_recap
+          required: false
+        players:
+          source: narratio.input.players
+          required: true
+        party:
+          source: narratio.input.party
+          required: true
+        glossary:
+          source: narratio.input.glossary
+          required: false
+      vars:
+        session_id: true
+        session_date: true
+        campaign_name: true
+        previous_session_id: true
+        output_kind: session_recap
+
+    # Example dependent artifact:
+    # - depends_on entries use artifact keys.
+    # - narratio.artifact.<key> sources require matching depends_on membership.
+    player_handout:
+      enabled: true
+      depends_on:
+        - session_recap
+      prompt_id: dnd.player_handout
+      profile_id: local-fast
+      output_path: artifacts/player_handout.md
+      timeout: 10m
+      inputs:
+        recap:
+          source: narratio.artifact.session_recap
+          required: true
+        transcript:
+          source: narratio.transcript.final_trimmed
+          required: true
+      vars:
+        session_id: true
+        campaign_name: true
+        output_kind: player_handout
+
+notification:
+  # No delivery provider is currently implemented.
+  mode: noop

examples/pipeline.minimal.yml

@@ -1,113 +1,6 @@
-workspace:
-  root: ./tmp/narratio-workspace
-
-storage:
-  backend: local
-
-secrets:
-  # Optional: load environment variables from files in this directory.
-  # File name = env var name; file contents = env var value.
-  env_dir: /var/local/narratio/secrets
+campaigns:
+  root: /usr/local/share/narratio/campaigns
+  default_campaign_id: sample-campaign
 
 whisperx:
   transcribe_url: "https://transcription.example.com/transcribe"
-  language: "en"
-  timeout: "30m"
-  retries: 3
-  retry_delay: "2s"
-  concurrency: 2
-
-seriatim:
-  binary: "seriatim"
-  timeout: "10m"
-  output_schema: "seriatim-intermediate"
-  coalesce_gap: 3.0
-  report: true
-  env:
-    overlap_word_run_gap: 1.0
-    overlap_word_run_reorder_window: 1.0
-    backchannel_max_duration: 2.0
-    filler_max_duration: 1.25
-
-audita:
-  binary: "audita"
-  timeout: "3h"
-  llm_api_key_env: "AUDITA_LLM_API_KEY"
-  # Optional: pass only when overriding Audita's default module sequence.
-  modules: []
-  base_url: "https://openrouter.ai/api/v1"
-  model: "openrouter/google/gemma-4-31b-it"
-  transcript_description: ""
-  config_path: ""
-  output_schema: "audita-v1"
-  work_dir_retention: "auto"
-  total_llm_concurrency: 1
-  proposal_llm_concurrency: 1
-  validation_model: ""
-  validation_llm_concurrency: 1
-  report: true
-
-normalize:
-  # Session-workdir-relative when not absolute.
-  output_path: "transcripts/normalized.json"
-  output_schema: "seriatim-intermediate"
-  report: true
-
-trim:
-  enabled: true
-  # Session-workdir-relative when not absolute.
-  output_path: "transcripts/trimmed.json"
-  bounds:
-    prompt_id: "dnd_session.bounds"
-    # Empty means use prompt default profile.
-    profile_id: ""
-    transcript_input_name: "transcript"
-    output_path: "artifacts/session_bounds.json"
-    timeout: "10m"
-    render_debug: false
-    render_output_path: "artifacts/session_bounds.render.json"
-  seriatim:
-    report: false
-
-scriptorium:
-  binary: "scriptorium"
-  config_path: "/etc/scriptorium/config.yml"
-  timeout: "10m"
-  render_debug: false
-  artifacts:
-    session_recap:
-      enabled: true
-      prompt_id: "dnd.session_recap"
-      profile_id: "local-quality"
-      output_path: "artifacts/session_recap.md"
-      timeout: "10m"
-      # Optional per-artifact override of global scriptorium.render_debug.
-      # render_debug: true
-      inputs:
-        transcript:
-          # Available transcript sources:
-          # - trimmed_transcript (recommended for session_recap)
-          # - normalized_transcript (recommended for future full-session analysis)
-          # - processed_transcript (raw Audita-polished output)
-          source: "trimmed_transcript"
-          required: true
-        previous_recap:
-          source: "previous_session_artifact"
-          artifact: "session_recap"
-          # Optional: set when previous recap is available.
-          path: ""
-          required: false
-      vars:
-        session_id: true
-        session_date: true
-        campaign_name: true
-        previous_session_id: true
-        output_kind: "session_recap"
-
-analyzer:
-  timeout: 20m
-  artifacts:
-    output_dir: artifacts
-
-notification:
-  timeout: 10s

examples/pipeline.production.yml

@@ -0,0 +1,128 @@
+workspace:
+  root: /var/lib/narratio/workspace
+  cleanup_after_publish: true
+
+storage:
+  backend: s3
+  s3:
+    bucket: my-dnd-archive
+    root_prefix: dnd
+    region: us-east-1
+    access_key_id_env: OBJECT_STORAGE_KEY_ID
+    secret_access_key_env: OBJECT_STORAGE_KEY
+
+campaigns:
+  root: /usr/local/share/narratio/campaigns
+  default_campaign_id: sample-campaign
+
+spool:
+  root: /var/spool/narratio
+  delete_audio_after_publish: true
+
+publish:
+  enabled: true
+  upload_run: true
+  outputs:
+    - source: narratio.transcript.final_trimmed
+      dest: transcripts/final.trimmed.json
+      required: true
+    - source: narratio.transcript.final_markdown
+      dest: transcripts/final.md
+      required: true
+    - source: narratio.transcript.final_trimmed_markdown
+      dest: transcripts/final.trimmed.md
+      required: true
+    - source: narratio.artifact.session_recap
+      dest: artifacts/session_recap.md
+      required: true
+    - source: narratio.artifact.player_handout
+      dest: artifacts/player_handout.md
+      required: false
+
+whisperx:
+  transcribe_url: "https://transcription.example.com/transcribe"
+  language: en
+  timeout: 45m
+  retries: 3
+  retry_delay: 3s
+  concurrency: 2
+
+seriatim:
+  binary: seriatim
+  timeout: 10m
+  output_schema: seriatim-intermediate
+  coalesce_gap: 3.0
+  report: true
+
+audita:
+  binary: audita
+  timeout: 3h
+  llm_api_key_env: AUDITA_LLM_API_KEY
+  modules: [glossary, homophones, spoken_word, grammar]
+  output_schema: audita-v1
+  work_dir_retention: auto
+  total_llm_concurrency: 2
+  proposal_llm_concurrency: 1
+  validation_llm_concurrency: 1
+  report: true
+
+normalize:
+  output_path: transcripts/final.json
+  output_schema: seriatim-intermediate
+  report: true
+
+scriptorium:
+  binary: scriptorium
+  config_path: /usr/local/etc/scriptorium/config.yml
+  timeout: 10m
+  render_debug: false
+  artifacts:
+    session_recap:
+      enabled: true
+      prompt_id: dnd.session_recap
+      profile_id: local-fast
+      output_path: artifacts/session_recap.md
+      timeout: 10m
+      inputs:
+        transcript:
+          source: narratio.transcript.final_trimmed
+          required: true
+        previous_recap:
+          source: narratio.previous_session.artifact.session_recap
+          required: false
+        players:
+          source: narratio.input.players
+          required: true
+        party:
+          source: narratio.input.party
+          required: true
+        glossary:
+          source: narratio.input.glossary
+          required: false
+      vars:
+        session_id: true
+        session_date: true
+        campaign_name: true
+        previous_session_id: true
+        output_kind: session_recap
+    player_handout:
+      enabled: true
+      depends_on:
+        - session_recap
+      prompt_id: dnd.player_handout
+      profile_id: local-fast
+      output_path: artifacts/player_handout.md
+      timeout: 10m
+      inputs:
+        recap:
+          source: narratio.artifact.session_recap
+          required: true
+        transcript:
+          source: narratio.transcript.final_trimmed
+          required: true
+      vars:
+        session_id: true
+        output_kind: player_handout
+
+notification:
+  mode: noop

examples/session.local-audio.yml

@@ -0,0 +1,5 @@
+session_id: 2026-05-03
+date: 2026-05-03
+title: Sample Session
+inputs:
+  audio_dir: ./audio

examples/session.minimal.yml

@@ -1,10 +0,0 @@
-session_id: 2026-05-03
-campaign: sample-campaign
-date: 2026-05-03
-title: Sample Session
-inputs:
-  audio_dir: ./audio
-  speakers_file: ./speakers.yml
-  autocorrect_file: ./autocorrect.yml
-  glossary_file: ./glossary.yml
-

examples/session.s3-audio.yml

@@ -0,0 +1,6 @@
+session_id: 2026-05-03
+date: 2026-05-03
+title: Sample Session
+inputs:
+  audio_s3:
+    prefix: audio/

examples/session.template.yml

@@ -0,0 +1,3 @@
+session_id: "{{ session_id }}"
+inputs:
+  audio_dir: ./audio

examples/speakers.yml

@@ -1,5 +1,5 @@
 match:
-  - speaker: "Eric Rakestraw"
+  - speaker: "Example Speaker"
     match:
-      - "Eric_Rakestraw"
-      - "Eric"
+      - "Example_Speaker"
+      - "Example"

go.mod

@@ -2,4 +2,28 @@ module gitea.maximumdirect.net/eric/narratio
 
 go 1.25.0
 
-require gopkg.in/yaml.v3 v3.0.1
+require (
+	github.com/aws/aws-sdk-go-v2/config v1.32.17
+	github.com/aws/aws-sdk-go-v2/credentials v1.19.16
+	github.com/aws/aws-sdk-go-v2/service/s3 v1.101.0
+	github.com/aws/smithy-go v1.25.1
+	golang.org/x/sys v0.47.0
+	gopkg.in/yaml.v3 v3.0.1
+)
+
+require (
+	github.com/aws/aws-sdk-go-v2 v1.41.7 // indirect
+	github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.10 // indirect
+	github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.18.23 // indirect
+	github.com/aws/aws-sdk-go-v2/internal/configsources v1.4.23 // indirect
+	github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.7.23 // indirect
+	github.com/aws/aws-sdk-go-v2/internal/v4a v1.4.24 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.9 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/checksum v1.9.15 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.13.23 // indirect
+	github.com/aws/aws-sdk-go-v2/service/internal/s3shared v1.19.23 // indirect
+	github.com/aws/aws-sdk-go-v2/service/signin v1.0.11 // indirect
+	github.com/aws/aws-sdk-go-v2/service/sso v1.30.17 // indirect
+	github.com/aws/aws-sdk-go-v2/service/ssooidc v1.35.21 // indirect
+	github.com/aws/aws-sdk-go-v2/service/sts v1.42.1 // indirect
+)

go.sum

@@ -1,3 +1,41 @@
+github.com/aws/aws-sdk-go-v2 v1.41.7 h1:DWpAJt66FmnnaRIOT/8ASTucrvuDPZASqhhLey6tLY8=
+github.com/aws/aws-sdk-go-v2 v1.41.7/go.mod h1:4LAfZOPHNVNQEckOACQx60Y8pSRjIkNZQz1w92xpMJc=
+github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.10 h1:gx1AwW1Iyk9Z9dD9F4akX5gnN3QZwUB20GGKH/I+Rho=
+github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.10/go.mod h1:qqY157uZoqm5OXq/amuaBJyC9hgBCBQnsaWnPe905GY=
+github.com/aws/aws-sdk-go-v2/config v1.32.17 h1:FpL4/758/diKwqbytU0prpuiu60fgXKUWCpDJtApclU=
+github.com/aws/aws-sdk-go-v2/config v1.32.17/go.mod h1:OXqUMzgXytfoF9JaKkhrOYsyh72t9G+MJH8mMRaexOE=
+github.com/aws/aws-sdk-go-v2/credentials v1.19.16 h1:r3RJBuU7X9ibt8RHbMjWE6y60QbKBiII6wSrXnapxSU=
+github.com/aws/aws-sdk-go-v2/credentials v1.19.16/go.mod h1:6cx7zqDENJDbBIIWX6P8s0h6hqHC8Avbjh9Dseo27ug=
+github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.18.23 h1:UuSfcORqNSz/ey3VPRS8TcVH2Ikf0/sC+Hdj400QI6U=
+github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.18.23/go.mod h1:+G/OSGiOFnSOkYloKj/9M35s74LgVAdJBSD5lsFfqKg=
+github.com/aws/aws-sdk-go-v2/internal/configsources v1.4.23 h1:GpT/TrnBYuE5gan2cZbTtvP+JlHsutdmlV2YfEyNde0=
+github.com/aws/aws-sdk-go-v2/internal/configsources v1.4.23/go.mod h1:xYWD6BS9ywC5bS3sz9Xh04whO/hzK2plt2Zkyrp4JuA=
+github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.7.23 h1:bpd8vxhlQi2r1hiueOw02f/duEPTMK59Q4QMAoTTtTo=
+github.com/aws/aws-sdk-go-v2/internal/endpoints/v2 v2.7.23/go.mod h1:15DfR2nw+CRHIk0tqNyifu3G1YdAOy68RftkhMDDwYk=
+github.com/aws/aws-sdk-go-v2/internal/v4a v1.4.24 h1:OQqn11BtaYv1WLUowvcA30MpzIu8Ti4pcLPIIyoKZrA=
+github.com/aws/aws-sdk-go-v2/internal/v4a v1.4.24/go.mod h1:X5ZJyfwVrWA96GzPmUCWFQaEARPR7gCrpq2E92PJwAE=
+github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.9 h1:FLudkZLt5ci0ozzgkVo8BJGwvqNaZbTWb3UcucAateA=
+github.com/aws/aws-sdk-go-v2/service/internal/accept-encoding v1.13.9/go.mod h1:w7wZ/s9qK7c8g4al+UyoF1Sp/Z45UwMGcqIzLWVQHWk=
+github.com/aws/aws-sdk-go-v2/service/internal/checksum v1.9.15 h1:ieLCO1JxUWuxTZ1cRd0GAaeX7O6cIxnwk7tc1LsQhC4=
+github.com/aws/aws-sdk-go-v2/service/internal/checksum v1.9.15/go.mod h1:e3IzZvQ3kAWNykvE0Tr0RDZCMFInMvhku3qNpcIQXhM=
+github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.13.23 h1:pbrxO/kuIwgEsOPLkaHu0O+m4fNgLU8B3vxQ+72jTPw=
+github.com/aws/aws-sdk-go-v2/service/internal/presigned-url v1.13.23/go.mod h1:/CMNUqoj46HpS3MNRDEDIwcgEnrtZlKRaHNaHxIFpNA=
+github.com/aws/aws-sdk-go-v2/service/internal/s3shared v1.19.23 h1:03xatSQO4+AM1lTAbnRg5OK528EUg744nW7F73U8DKw=
+github.com/aws/aws-sdk-go-v2/service/internal/s3shared v1.19.23/go.mod h1:M8l3mwgx5ToK7wot2sBBce/ojzgnPzZXUV445gTSyE8=
+github.com/aws/aws-sdk-go-v2/service/s3 v1.101.0 h1:etqBTKY581iwLL/H/S2sVgk3C9lAsTJFeXWFDsDcWOU=
+github.com/aws/aws-sdk-go-v2/service/s3 v1.101.0/go.mod h1:L2dcoOgS2VSgbPLvpak2NyUPsO1TBN7M45Z4H7DlRc4=
+github.com/aws/aws-sdk-go-v2/service/signin v1.0.11 h1:TdJ+HdzOBhU8+iVAOGUTU63VXopcumCOF1paFulHWZc=
+github.com/aws/aws-sdk-go-v2/service/signin v1.0.11/go.mod h1:R82ZRExE/nheo0N+T8zHPcLRTcH8MGsnR3BiVGX0TwI=
+github.com/aws/aws-sdk-go-v2/service/sso v1.30.17 h1:7byT8HUWrgoRp6sXjxtZwgOKfhss5fW6SkLBtqzgRoE=
+github.com/aws/aws-sdk-go-v2/service/sso v1.30.17/go.mod h1:xNWknVi4Ezm1vg1QsB/5EWpAJURq22uqd38U8qKvOJc=
+github.com/aws/aws-sdk-go-v2/service/ssooidc v1.35.21 h1:+1Kl1zx6bWi4X7cKi3VYh29h8BvsCoHQEQ6ST9X8w7w=
+github.com/aws/aws-sdk-go-v2/service/ssooidc v1.35.21/go.mod h1:4vIRDq+CJB2xFAXZ+YgGUTiEft7oAQlhIs71xcSeuVg=
+github.com/aws/aws-sdk-go-v2/service/sts v1.42.1 h1:F/M5Y9I3nwr2IEpshZgh1GeHpOItExNM9L1euNuh/fk=
+github.com/aws/aws-sdk-go-v2/service/sts v1.42.1/go.mod h1:mTNxImtovCOEEuD65mKW7DCsL+2gjEH+RPEAexAzAio=
+github.com/aws/smithy-go v1.25.1 h1:J8ERsGSU7d+aCmdQur5Txg6bVoYelvQJgtZehD12GkI=
+github.com/aws/smithy-go v1.25.1/go.mod h1:YE2RhdIuDbA5E5bTdciG9KrW3+TiEONeUWCqxX9i1Fc=
+golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=
+golang.org/x/sys v0.47.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=

internal/adapters/analyzer/fake.go

@@ -1,40 +0,0 @@
-package analyzer
-
-import "context"
-
-// NoopRunner is a deterministic no-op analyzer adapter.
-type NoopRunner struct{}
-
-// Run returns the requested output path with placeholder metadata.
-func (n *NoopRunner) Run(ctx context.Context, req AnalyzeRequest) (AnalyzeResult, error) {
-	if err := ctx.Err(); err != nil {
-		return AnalyzeResult{}, err
-	}
-	return AnalyzeResult{ArtifactPath: req.OutputPath, Metadata: map[string]any{"placeholder": true}}, nil
-}
-
-// FakeRunner captures analyze requests and returns deterministic responses.
-type FakeRunner struct {
-	Requests []AnalyzeRequest
-	Err      error
-	Result   AnalyzeResult
-}
-
-// Run records request and returns configured response.
-func (f *FakeRunner) Run(ctx context.Context, req AnalyzeRequest) (AnalyzeResult, error) {
-	if err := ctx.Err(); err != nil {
-		return AnalyzeResult{}, err
-	}
-	f.Requests = append(f.Requests, req)
-	if f.Err != nil {
-		return AnalyzeResult{}, f.Err
-	}
-	res := f.Result
-	if res.ArtifactPath == "" {
-		res.ArtifactPath = req.OutputPath
-	}
-	if res.Metadata == nil {
-		res.Metadata = map[string]any{"fake": true}
-	}
-	return res, nil
-}

internal/adapters/analyzer/fake_test.go

@@ -1,31 +0,0 @@
-package analyzer
-
-import (
-	"context"
-	"errors"
-	"testing"
-)
-
-func TestFakeRunnerCapturesRequestAndReturnsPath(t *testing.T) {
-	fake := &FakeRunner{}
-	req := AnalyzeRequest{ArtifactType: "session-log", OutputPath: "artifacts/session-log.md"}
-
-	res, err := fake.Run(context.Background(), req)
-	if err != nil {
-		t.Fatalf("Run() error = %v", err)
-	}
-	if len(fake.Requests) != 1 || fake.Requests[0].ArtifactType != "session-log" {
-		t.Fatalf("requests = %#v, want captured request", fake.Requests)
-	}
-	if res.ArtifactPath != req.OutputPath {
-		t.Fatalf("artifact path = %q, want %q", res.ArtifactPath, req.OutputPath)
-	}
-}
-
-func TestFakeRunnerError(t *testing.T) {
-	fake := &FakeRunner{Err: errors.New("boom")}
-	_, err := fake.Run(context.Background(), AnalyzeRequest{})
-	if err == nil {
-		t.Fatal("expected error, got nil")
-	}
-}

internal/adapters/analyzer/runner.go

@@ -1,28 +0,0 @@
-// Package analyzer declares the adapter contract for artifact analysis generation.
-package analyzer
-
-import "context"
-
-// TODO: implement analyzer integration once the analyzer contract is finalized.
-
-// Runner is the adapter boundary for analyzer invocations.
-type Runner interface {
-	Run(ctx context.Context, req AnalyzeRequest) (AnalyzeResult, error)
-}
-
-// AnalyzeRequest describes one analyzer artifact generation request.
-type AnalyzeRequest struct {
-	ArtifactType            string
-	ProcessedTranscriptPath string
-	ContextReferences       []string
-	OutputPath              string
-	GeneratedConfigPath     string
-	StdoutLogPath           string
-	StderrLogPath           string
-}
-
-// AnalyzeResult describes analyzer output.
-type AnalyzeResult struct {
-	ArtifactPath string
-	Metadata     map[string]any
-}

internal/adapters/audita/fake.go

@@ -4,10 +4,10 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
-	"os"
 	"path/filepath"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
 )
 
 // NoopRunner is a deterministic no-op audita adapter.
@@ -84,17 +84,17 @@ func materializePlaceholders(req PolishRequest) error {
 			"merged_transcript_path": req.MergedTranscriptPath,
 			"output_path":            req.OutputProcessedPath,
 		}
-		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644); err != nil {
+		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 	if req.StdoutLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("audita noop/fake stdout placeholder\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("audita noop/fake stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
 		}
 	}
 	if req.StderrLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("audita noop/fake stderr placeholder\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("audita noop/fake stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
 		}
 	}
@@ -117,14 +117,14 @@ func writeJSONIfRequested(path string, payload any) error {
 	if path == "" {
 		return nil
 	}
-	if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
+	if err := fileops.EnsureWorkspaceDirectory(filepath.Dir(path)); err != nil {
 		return fmt.Errorf("create parent directory %q: %w", filepath.Dir(path), err)
 	}
 	data, err := json.Marshal(payload)
 	if err != nil {
 		return fmt.Errorf("marshal placeholder json for %q: %w", path, err)
 	}
-	if err := subprocess.WriteFileAtomic(path, data, 0o644); err != nil {
+	if err := subprocess.WriteFileAtomic(path, data, fileops.WorkspaceFileMode); err != nil {
 		return fmt.Errorf("write placeholder json %q: %w", path, err)
 	}
 	return nil

internal/adapters/audita/fake_test.go

@@ -14,7 +14,7 @@ func TestFakeRunnerCapturesRequestAndReturnsPath(t *testing.T) {
 	dir := t.TempDir()
 	req := PolishRequest{
 		GeneratedConfigPath: filepath.Join(dir, "config", "audita.yml"),
-		OutputProcessedPath: filepath.Join(dir, "transcripts", "processed.json"),
+		OutputProcessedPath: filepath.Join(dir, "transcripts", "polished.json"),
 		StdoutLogPath:       filepath.Join(dir, "logs", "audita.stdout.log"),
 		StderrLogPath:       filepath.Join(dir, "logs", "audita.stderr.log"),
 	}

internal/adapters/audita/runner.go

@@ -6,8 +6,6 @@ import (
 	"time"
 )
 
-// TODO: implement a real Audita subprocess/service adapter.
-
 // Runner is the adapter boundary for audita polish invocations.
 type Runner interface {
 	Run(ctx context.Context, req PolishRequest) (PolishResult, error)
@@ -15,25 +13,15 @@ type Runner interface {
 
 // PolishRequest describes an audita invocation.
 type PolishRequest struct {
-	GeneratedConfigPath      string
-	MergedTranscriptPath     string
-	OutputProcessedPath      string
-	GlossaryPath             string
-	ReportPath               string
-	WorkDir                  string
-	Modules                  []string
-	BaseURL                  string
-	Model                    string
-	TranscriptDescription    string
-	ConfigPath               string
-	OutputSchema             string
-	WorkDirRetention         string
-	TotalLLMConcurrency      *int
-	ProposalLLMConcurrency   *int
-	ValidationModel          string
-	ValidationLLMConcurrency *int
-	StdoutLogPath            string
-	StderrLogPath            string
+	GeneratedConfigPath  string
+	MergedTranscriptPath string
+	OutputProcessedPath  string
+	GlossaryPath         string
+	ReportPath           string
+	WorkDir              string
+	Modules              []string
+	StdoutLogPath        string
+	StderrLogPath        string
 }
 
 // PolishResult describes a polish output.

internal/adapters/audita/subprocess.go

@@ -11,8 +11,15 @@ import (
 	"time"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
 )
 
+// MaxProcessedOutputBytes bounds Audita's processed-transcript JSON result.
+const MaxProcessedOutputBytes int64 = 64 * 1024 * 1024
+
+// MaxReportOutputBytes bounds Audita's optional report JSON result.
+const MaxReportOutputBytes int64 = 16 * 1024 * 1024
+
 // SubprocessRunnerConfig defines deterministic settings for Audita CLI execution.
 type SubprocessRunnerConfig struct {
 	Binary                   string
@@ -108,18 +115,14 @@ func NewSubprocessRunner(cfg SubprocessRunnerConfig) (*SubprocessRunner, error)
 			return nil, fmt.Errorf("audita module at index %d is empty", i)
 		}
 	}
-	if strings.TrimSpace(cfg.BaseURL) == "" {
-		return nil, fmt.Errorf("audita base url is required")
-	}
-	u, err := url.Parse(cfg.BaseURL)
-	if err != nil || u.Scheme == "" || u.Host == "" {
-		if err != nil {
-			return nil, fmt.Errorf("audita base url %q is invalid: %w", cfg.BaseURL, err)
+	if strings.TrimSpace(cfg.BaseURL) != "" {
+		u, err := url.Parse(cfg.BaseURL)
+		if err != nil || u.Scheme == "" || u.Host == "" {
+			if err != nil {
+				return nil, fmt.Errorf("audita base url %q is invalid: %w", cfg.BaseURL, err)
+			}
+			return nil, fmt.Errorf("audita base url %q is invalid", cfg.BaseURL)
 		}
-		return nil, fmt.Errorf("audita base url %q is invalid", cfg.BaseURL)
-	}
-	if strings.TrimSpace(cfg.Model) == "" {
-		return nil, fmt.Errorf("audita model is required")
 	}
 	if cfg.TotalLLMConcurrency != nil && *cfg.TotalLLMConcurrency <= 0 {
 		return nil, fmt.Errorf("audita total llm concurrency must be > 0 when provided")
@@ -211,12 +214,13 @@ func (r *SubprocessRunner) Run(ctx context.Context, req PolishRequest) (PolishRe
 	}
 
 	runRes, err := subprocess.Run(ctx, subprocess.RunRequest{
-		Executable:    r.binary,
-		Args:          args,
-		Timeout:       r.timeout,
-		EnvOverrides:  env,
-		StdoutLogPath: req.StdoutLogPath,
-		StderrLogPath: req.StderrLogPath,
+		Executable:      r.binary,
+		Args:            args,
+		Timeout:         r.timeout,
+		EnvOverrides:    env,
+		DiagnosticOwner: "audita",
+		StdoutLogPath:   req.StdoutLogPath,
+		StderrLogPath:   req.StderrLogPath,
 	})
 	if err != nil {
 		wrappedMessage := fmt.Sprintf(
@@ -307,10 +311,14 @@ func (r *SubprocessRunner) buildArgs(req PolishRequest, modules []string) []stri
 		req.MergedTranscriptPath,
 		"--glossary", req.GlossaryPath,
 		"--output", req.OutputProcessedPath,
-		"--base-url", r.baseURL,
-		"--model", r.model,
 		"--work-dir", req.WorkDir,
 	}
+	if r.baseURL != "" {
+		args = append(args, "--base-url", r.baseURL)
+	}
+	if r.model != "" {
+		args = append(args, "--model", r.model)
+	}
 	if len(modules) > 0 {
 		args = append(args, "--modules", strings.Join(modules, ","))
 	}
@@ -370,13 +378,13 @@ func (r *SubprocessRunner) writeInvocationConfig(req PolishRequest, args []strin
 		"credential_env_var":         r.llmAPIKeyEnv,
 		"credential_present":         credentialPresent,
 	}
-	return subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644)
+	return subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, fileops.WorkspaceFileMode)
 }
 
 func validateProcessedOutput(path string) error {
-	data, err := os.ReadFile(path)
+	data, err := readAuditaResult(path, MaxProcessedOutputBytes, "processed transcript")
 	if err != nil {
-		return fmt.Errorf("read file: %w", err)
+		return err
 	}
 
 	var payload map[string]any
@@ -405,9 +413,9 @@ func addSubprocessStreamHint(message string, runErr error) string {
 }
 
 func validateJSONFile(path string) error {
-	data, err := os.ReadFile(path)
+	data, err := readAuditaResult(path, MaxReportOutputBytes, "report")
 	if err != nil {
-		return fmt.Errorf("read file: %w", err)
+		return err
 	}
 	var v any
 	if err := json.Unmarshal(data, &v); err != nil {
@@ -415,3 +423,11 @@ func validateJSONFile(path string) error {
 	}
 	return nil
 }
+
+func readAuditaResult(path string, limit int64, category string) ([]byte, error) {
+	data, err := fileops.ReadRegularFile(path, limit)
+	if err != nil {
+		return nil, fmt.Errorf("audita %s result exceeds or cannot be read within %d-byte limit: %w", category, limit, err)
+	}
+	return data, nil
+}

internal/adapters/audita/subprocess_test.go

@@ -52,9 +52,9 @@ func TestSubprocessRunnerSuccessArgsEnvAndValidation(t *testing.T) {
 	dir := t.TempDir()
 	req := PolishRequest{
 		GeneratedConfigPath:  filepath.Join(dir, "audita.generated.yml"),
-		MergedTranscriptPath: filepath.Join(dir, "merged.json"),
+		MergedTranscriptPath: filepath.Join(dir, "base.json"),
 		GlossaryPath:         filepath.Join(dir, "glossary.yml"),
-		OutputProcessedPath:  filepath.Join(dir, "processed.json"),
+		OutputProcessedPath:  filepath.Join(dir, "polished.json"),
 		ReportPath:           filepath.Join(dir, "audita.report.json"),
 		WorkDir:              filepath.Join(dir, "artifacts", "audita-work"),
 		StdoutLogPath:        filepath.Join(dir, "audita.stdout.log"),
@@ -99,9 +99,9 @@ func TestSubprocessRunnerSuccessArgsEnvAndValidation(t *testing.T) {
 		"process", req.MergedTranscriptPath,
 		"--glossary", req.GlossaryPath,
 		"--output", req.OutputProcessedPath,
+		"--work-dir", req.WorkDir,
 		"--base-url", "https://openrouter.ai/api/v1",
 		"--model", "openrouter/google/gemma-4-31b-it",
-		"--work-dir", req.WorkDir,
 		"--modules", "glossary,homophones,glossary",
 		"--report-json", req.ReportPath,
 		"--transcript-description", "Campaign Session 42",
@@ -189,7 +189,7 @@ func TestSubprocessRunnerUnconfiguredCredentialEnvOmitsCredential(t *testing.T)
 	}
 }
 
-func TestSubprocessRunnerInheritsParentEnvironment(t *testing.T) {
+func TestSubprocessRunnerOmitsUnspecifiedParentEnvironment(t *testing.T) {
 	if runtime.GOOS == "windows" {
 		t.Skip("helper wrapper script uses /bin/sh")
 	}
@@ -214,8 +214,8 @@ func TestSubprocessRunnerInheritsParentEnvironment(t *testing.T) {
 	}
 
 	rec := readAuditaHelperRecord(t, recordPath)
-	if rec.Env["AUDITA_INHERITED_MARKER"] != "inherited-from-parent" {
-		t.Fatalf("AUDITA_INHERITED_MARKER = %q, want inherited-from-parent", rec.Env["AUDITA_INHERITED_MARKER"])
+	if rec.Env["AUDITA_INHERITED_MARKER"] != "" {
+		t.Fatalf("AUDITA_INHERITED_MARKER = %q, want omitted from the child environment", rec.Env["AUDITA_INHERITED_MARKER"])
 	}
 }
 
@@ -248,6 +248,36 @@ func TestSubprocessRunnerOmitsModulesFlagWhenNotConfigured(t *testing.T) {
 	}
 }
 
+func TestSubprocessRunnerOmitsBaseURLAndModelFlagsWhenNotConfigured(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("helper wrapper script uses /bin/sh")
+	}
+	t.Setenv("GO_WANT_AUDITA_HELPER", "1")
+	t.Setenv("AUDITA_HELPER_MODE", "success")
+	recordPath := filepath.Join(t.TempDir(), "record.json")
+	t.Setenv("AUDITA_HELPER_RECORD_PATH", recordPath)
+
+	runner := mustAuditaRunner(t, SubprocessRunnerConfig{
+		Binary:       writeAuditaHelperWrapper(t),
+		Timeout:      mustParseAuditaDuration(t, "2s"),
+		LLMAPIKeyEnv: "",
+		Report:       false,
+	})
+	req := auditaReqForTest(t, false)
+	if _, err := runner.Run(context.Background(), req); err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	rec := readAuditaHelperRecord(t, recordPath)
+	for i := 0; i < len(rec.Args); i++ {
+		if rec.Args[i] == "--base-url" {
+			t.Fatalf("args contained --base-url unexpectedly: %#v", rec.Args)
+		}
+		if rec.Args[i] == "--model" {
+			t.Fatalf("args contained --model unexpectedly: %#v", rec.Args)
+		}
+	}
+}
+
 func TestSubprocessRunnerSubprocessFailure(t *testing.T) {
 	if runtime.GOOS == "windows" {
 		t.Skip("helper wrapper script uses /bin/sh")
@@ -541,7 +571,7 @@ func mustAuditaRunner(t *testing.T, cfg SubprocessRunnerConfig) *SubprocessRunne
 func auditaReqForTest(t *testing.T, withReport bool) PolishRequest {
 	t.Helper()
 	dir := t.TempDir()
-	merged := filepath.Join(dir, "merged.json")
+	merged := filepath.Join(dir, "base.json")
 	glossary := filepath.Join(dir, "glossary.yml")
 	writeAuditaTestFile(t, merged, `{"segments":[]}`)
 	writeAuditaTestFile(t, glossary, "terms: []\n")
@@ -549,7 +579,7 @@ func auditaReqForTest(t *testing.T, withReport bool) PolishRequest {
 		GeneratedConfigPath:  filepath.Join(dir, "audita.generated.yml"),
 		MergedTranscriptPath: merged,
 		GlossaryPath:         glossary,
-		OutputProcessedPath:  filepath.Join(dir, "processed.json"),
+		OutputProcessedPath:  filepath.Join(dir, "polished.json"),
 		WorkDir:              filepath.Join(dir, "artifacts", "audita-work"),
 		StdoutLogPath:        filepath.Join(dir, "audita.stdout.log"),
 		StderrLogPath:        filepath.Join(dir, "audita.stderr.log"),

internal/adapters/notarius/fake.go

@@ -0,0 +1,24 @@
+package notarius
+
+import "context"
+
+// FakeRunner is a configurable in-memory runner for stage tests.
+type FakeRunner struct {
+	Requests []RunRequest
+	Result   RunResult
+	Err      error
+}
+
+// Run records the request and returns the configured result or error.
+func (f *FakeRunner) Run(ctx context.Context, req RunRequest) (RunResult, error) {
+	if err := ctx.Err(); err != nil {
+		return RunResult{}, err
+	}
+	copyRequest := req
+	copyRequest.References = append([]ReferenceBinding(nil), req.References...)
+	f.Requests = append(f.Requests, copyRequest)
+	if f.Err != nil {
+		return RunResult{}, f.Err
+	}
+	return f.Result, nil
+}

internal/adapters/notarius/runner.go

@@ -0,0 +1,160 @@
+// Package notarius declares the adapter contract for Notarius CLI invocations.
+package notarius
+
+import (
+	"context"
+	"time"
+)
+
+const ReceiptSchemaVersion = "notarius.run-result.v2"
+
+// Runner is the adapter boundary for a complete Notarius pipeline invocation.
+type Runner interface {
+	Run(ctx context.Context, req RunRequest) (RunResult, error)
+}
+
+// ReferenceBinding maps one normalized Notarius selector to an absolute
+// external reference path.
+type ReferenceBinding struct {
+	Selector string
+	Path     string
+}
+
+// RunRequest contains the resolved inputs and diagnostic destinations for one invocation.
+type RunRequest struct {
+	Binary           string
+	ConfigPath       string
+	PipelineID       string
+	InputPath        string
+	OutputRoot       string
+	WorkingDirectory string
+	ReceiptPath      string
+	LogPath          string
+	Timeout          time.Duration
+	References       []ReferenceBinding
+}
+
+// Receipt is the transport-neutral successful run receipt.
+type Receipt struct {
+	SchemaVersion             string
+	RunID                     string
+	PipelineID                string
+	OutputDirectory           string
+	IndexFile                 string
+	NormalizedOutputCount     int
+	RejectedOutputCount       int
+	WarningGroupCount         int
+	WarningOccurrenceCount    int
+	DiagnosticGroupCount      int
+	DiagnosticOccurrenceCount int
+	DiagnosticsTruncated      bool
+	ValidationStatus          string
+	ValidationSummaries       []ValidationSummary
+	DebugDirectory            string
+}
+
+// ValidationSummary retains the bounded outcome of one Notarius producer result.
+type ValidationSummary struct {
+	Stage                string
+	StepID               string
+	LaneID               string
+	ModuleKey            string
+	ChunkID              string
+	Status               string
+	RejectingValidators  []string
+	ReasonCodes          []string
+	IncompleteValidators []string
+	ProducerAttemptCount int
+	TerminalAction       string
+}
+
+// LaneDescriptor identifies one normalized lane payload discovered through the index.
+type LaneDescriptor struct {
+	LaneID        string
+	File          string
+	Path          string
+	MediaType     string
+	ModuleKey     string
+	SchemaID      string
+	SchemaName    string
+	SchemaVersion string
+}
+
+// PipelineDescriptor identifies a pipeline-wide artifact discovered through the index.
+type PipelineDescriptor struct {
+	ArtifactKind  string
+	File          string
+	Path          string
+	MediaType     string
+	SchemaID      string
+	SchemaName    string
+	SchemaVersion string
+}
+
+// Index describes the validated bundle-management and artifact paths.
+type Index struct {
+	Path            string
+	ManifestFile    string
+	ManifestPath    string
+	RejectedFile    string
+	RejectedPath    string
+	WarningsFile    string
+	WarningsPath    string
+	DiagnosticsFile string
+	DiagnosticsPath string
+	Lanes           []LaneDescriptor
+	ChunkMap        *PipelineDescriptor
+	EvidenceContext *PipelineDescriptor
+}
+
+// RejectionSummary retains structured rejection identity without free-form messages.
+type RejectionSummary struct {
+	Stage         string
+	StepID        string
+	LaneID        string
+	ModuleKey     string
+	ChunkID       string
+	ValidatorName string
+	ReasonCode    string
+}
+
+// WarningSummary retains structured warning identity without free-form messages.
+type WarningSummary struct {
+	Disposition     string
+	Category        string
+	ReasonCode      string
+	Origin          DiagnosticOrigin
+	OccurrenceCount int
+}
+
+// DiagnosticOrigin identifies the framework-owned pipeline location of a finding.
+type DiagnosticOrigin struct {
+	Stage        string
+	StepID       string
+	LaneID       string
+	ModuleKey    string
+	ValidatorKey string
+}
+
+// DiagnosticSummary retains bounded advisory or observation group metadata.
+type DiagnosticSummary struct {
+	Disposition     string
+	Category        string
+	ReasonCode      string
+	Origin          DiagnosticOrigin
+	OccurrenceCount int
+}
+
+// RunResult describes a successfully decoded and validated Notarius bundle.
+type RunResult struct {
+	Receipt     Receipt
+	Index       Index
+	BundleRoot  string
+	ReceiptPath string
+	LogPath     string
+	ExitCode    int
+	Duration    time.Duration
+	Rejections  []RejectionSummary
+	Warnings    []WarningSummary
+	Diagnostics []DiagnosticSummary
+}

internal/adapters/notarius/subprocess.go

@@ -0,0 +1,784 @@
+package notarius
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
+	"gitea.maximumdirect.net/eric/narratio/internal/notariusref"
+	"gitea.maximumdirect.net/eric/narratio/internal/pathsafe"
+)
+
+const (
+	maxReceiptBytes          = 1 << 20
+	maxIndexBytes            = 4 << 20
+	maxSummaryBytes          = 4 << 20
+	canonicalIndexFile       = "index.json"
+	canonicalManifestFile    = "manifest.json"
+	canonicalRejectedFile    = "rejected.json"
+	canonicalWarningsFile    = "warnings.json"
+	canonicalDiagnosticsFile = "diagnostics.json"
+	warningsSchemaVersion    = "notarius.warnings.v2"
+	diagnosticsSchemaVersion = "notarius.diagnostics.v1"
+	maxWarningGroups         = 128
+	maxDiagnosticGroups      = 256
+	maxFindingSamples        = 3
+)
+
+type subprocessRun func(context.Context, subprocess.RunRequest) (subprocess.RunResult, error)
+
+// SubprocessRunner invokes Notarius through its public CLI.
+type SubprocessRunner struct {
+	run subprocessRun
+}
+
+// NewSubprocessRunner constructs a production Notarius subprocess runner.
+func NewSubprocessRunner() *SubprocessRunner {
+	return &SubprocessRunner{run: subprocess.Run}
+}
+
+// Run executes a complete Notarius pipeline and discovers its published bundle.
+func (r *SubprocessRunner) Run(ctx context.Context, req RunRequest) (RunResult, error) {
+	if r == nil || r.run == nil {
+		return RunResult{}, fmt.Errorf("notarius subprocess runner is nil")
+	}
+	references, err := validateRunRequest(req)
+	if err != nil {
+		return RunResult{}, err
+	}
+
+	args := []string{
+		"run", req.PipelineID,
+		"--config", req.ConfigPath,
+		"--input", req.InputPath,
+		"--output-dir", req.OutputRoot,
+	}
+	for _, reference := range references {
+		args = append(args, "--reference", reference.Selector+"="+reference.Path)
+	}
+	args = append(args, "--json")
+	processResult, err := r.run(ctx, subprocess.RunRequest{
+		Executable:      req.Binary,
+		Args:            args,
+		WorkingDir:      req.WorkingDirectory,
+		Timeout:         req.Timeout,
+		DiagnosticOwner: "notarius",
+		StdoutLogPath:   req.ReceiptPath,
+		StderrLogPath:   req.LogPath,
+	})
+	baseResult := RunResult{
+		ReceiptPath: req.ReceiptPath,
+		LogPath:     req.LogPath,
+		ExitCode:    processResult.ExitCode,
+		Duration:    processResult.Duration,
+	}
+	if err != nil {
+		return baseResult, fmt.Errorf("run notarius pipeline %q: %w", req.PipelineID, err)
+	}
+
+	receipt, err := loadReceipt(req.ReceiptPath, req.PipelineID)
+	if err != nil {
+		return baseResult, err
+	}
+	bundleRoot, err := validateBundleRoot(req.OutputRoot, receipt.OutputDirectory)
+	if err != nil {
+		return baseResult, err
+	}
+	indexPath, err := resolveRegularFile(bundleRoot, receipt.IndexFile)
+	if err != nil {
+		return baseResult, fmt.Errorf("resolve receipt index file: %w", err)
+	}
+	index, err := loadIndex(bundleRoot, indexPath)
+	if err != nil {
+		return baseResult, err
+	}
+	rejections, err := loadRejections(index.RejectedPath)
+	if err != nil {
+		return baseResult, err
+	}
+	warnings, err := loadWarnings(index.WarningsPath)
+	if err != nil {
+		return baseResult, err
+	}
+	diagnostics, diagnosticOccurrences, diagnosticsTruncated, err := loadDiagnostics(index.DiagnosticsPath)
+	if err != nil {
+		return baseResult, err
+	}
+	if receipt.NormalizedOutputCount != len(index.Lanes) || receipt.RejectedOutputCount != len(rejections) ||
+		receipt.WarningGroupCount != len(warnings) || receipt.DiagnosticGroupCount != len(diagnostics) {
+		return baseResult, fmt.Errorf("notarius receipt counts do not match published bundle")
+	}
+	warningOccurrences, err := sumWarningOccurrences(warnings)
+	if err != nil {
+		return baseResult, err
+	}
+	if receipt.WarningOccurrenceCount != warningOccurrences ||
+		receipt.DiagnosticOccurrenceCount != diagnosticOccurrences ||
+		receipt.DiagnosticsTruncated != diagnosticsTruncated {
+		return baseResult, fmt.Errorf("notarius receipt occurrence counts do not match published bundle")
+	}
+
+	baseResult.Receipt = receipt
+	baseResult.Index = index
+	baseResult.BundleRoot = bundleRoot
+	baseResult.Rejections = rejections
+	baseResult.Warnings = warnings
+	baseResult.Diagnostics = diagnostics
+	return baseResult, nil
+}
+
+func validateRunRequest(req RunRequest) ([]ReferenceBinding, error) {
+	if strings.TrimSpace(req.Binary) == "" {
+		return nil, fmt.Errorf("notarius binary is required")
+	}
+	if strings.TrimSpace(req.PipelineID) == "" {
+		return nil, fmt.Errorf("notarius pipeline id is required")
+	}
+	if req.Timeout <= 0 {
+		return nil, fmt.Errorf("notarius timeout must be positive")
+	}
+	for label, path := range map[string]string{
+		"config":            req.ConfigPath,
+		"input":             req.InputPath,
+		"output root":       req.OutputRoot,
+		"working directory": req.WorkingDirectory,
+		"receipt":           req.ReceiptPath,
+		"log":               req.LogPath,
+	} {
+		if strings.TrimSpace(path) == "" {
+			return nil, fmt.Errorf("notarius %s path is required", label)
+		}
+		if !filepath.IsAbs(path) {
+			return nil, fmt.Errorf("notarius %s path must be absolute", label)
+		}
+	}
+	if filepath.Clean(req.ReceiptPath) == filepath.Clean(req.LogPath) {
+		return nil, fmt.Errorf("notarius receipt and log paths must be different")
+	}
+	references := make([]ReferenceBinding, 0, len(req.References))
+	selectors := make(map[string]struct{}, len(req.References))
+	for index, binding := range req.References {
+		selector, err := notariusref.NormalizeSelector(binding.Selector)
+		if err != nil {
+			return nil, fmt.Errorf("notarius reference %d selector: %w", index, err)
+		}
+		if _, duplicate := selectors[selector]; duplicate {
+			return nil, fmt.Errorf("notarius reference selector %q is duplicated", selector)
+		}
+		selectors[selector] = struct{}{}
+		if strings.TrimSpace(binding.Path) == "" {
+			return nil, fmt.Errorf("notarius reference %q path is required", selector)
+		}
+		if !filepath.IsAbs(binding.Path) {
+			return nil, fmt.Errorf("notarius reference %q path must be absolute", selector)
+		}
+		references = append(references, ReferenceBinding{Selector: selector, Path: binding.Path})
+	}
+	if err := requireRegularFile(req.ConfigPath); err != nil {
+		return nil, fmt.Errorf("validate notarius config path: %w", err)
+	}
+	if err := requireRegularFile(req.InputPath); err != nil {
+		return nil, fmt.Errorf("validate notarius input path: %w", err)
+	}
+	if err := requireDirectory(req.OutputRoot); err != nil {
+		return nil, fmt.Errorf("validate notarius output root: %w", err)
+	}
+	if err := requireDirectory(req.WorkingDirectory); err != nil {
+		return nil, fmt.Errorf("validate notarius working directory: %w", err)
+	}
+	if err := validateLogDestination(req.ReceiptPath); err != nil {
+		return nil, fmt.Errorf("validate notarius receipt path: %w", err)
+	}
+	if err := validateLogDestination(req.LogPath); err != nil {
+		return nil, fmt.Errorf("validate notarius log path: %w", err)
+	}
+	return references, nil
+}
+
+type receiptDocument struct {
+	SchemaVersion             string                      `json:"schema_version"`
+	RunID                     string                      `json:"run_id"`
+	PipelineID                string                      `json:"pipeline_id"`
+	OutputDirectory           string                      `json:"output_directory"`
+	IndexFile                 string                      `json:"index_file"`
+	NormalizedOutputCount     *int                        `json:"normalized_output_count"`
+	RejectedOutputCount       *int                        `json:"rejected_output_count"`
+	WarningGroupCount         *int                        `json:"warning_group_count"`
+	WarningOccurrenceCount    *int                        `json:"warning_occurrence_count"`
+	DiagnosticGroupCount      *int                        `json:"diagnostic_group_count"`
+	DiagnosticOccurrenceCount *int                        `json:"diagnostic_occurrence_count"`
+	DiagnosticsTruncated      *bool                       `json:"diagnostics_truncated"`
+	ValidationStatus          string                      `json:"validation_status"`
+	ValidationSummaries       []validationSummaryDocument `json:"validation_summaries"`
+	DebugDirectory            string                      `json:"debug_directory"`
+}
+
+type validationSummaryDocument struct {
+	Stage                string   `json:"stage"`
+	StepID               string   `json:"step_id"`
+	LaneID               string   `json:"lane_id"`
+	ModuleKey            string   `json:"module_key"`
+	ChunkID              string   `json:"chunk_id"`
+	Status               string   `json:"status"`
+	RejectingValidators  []string `json:"rejecting_validators"`
+	ReasonCodes          []string `json:"reason_codes"`
+	IncompleteValidators []string `json:"incomplete_validators"`
+	ProducerAttemptCount *int     `json:"producer_attempt_count"`
+	TerminalAction       string   `json:"terminal_action"`
+}
+
+func validValidationStatus(value string) bool {
+	switch value {
+	case "approved", "rejected", "incomplete":
+		return true
+	default:
+		return false
+	}
+}
+
+func validateValidationSummaries(documents []validationSummaryDocument) ([]ValidationSummary, error) {
+	summaries := make([]ValidationSummary, 0, len(documents))
+	for _, document := range documents {
+		if document.Status != "complete" && document.Status != "rejected" && document.Status != "incomplete" {
+			return nil, fmt.Errorf("notarius validation summary status %q is invalid", document.Status)
+		}
+		if document.ProducerAttemptCount == nil || *document.ProducerAttemptCount <= 0 || !validTerminalAction(document.TerminalAction) {
+			return nil, fmt.Errorf("notarius validation summary is missing required fields")
+		}
+		summaries = append(summaries, ValidationSummary{
+			Stage: document.Stage, StepID: document.StepID, LaneID: document.LaneID,
+			ModuleKey: document.ModuleKey, ChunkID: document.ChunkID, Status: document.Status,
+			RejectingValidators:  append([]string(nil), document.RejectingValidators...),
+			ReasonCodes:          append([]string(nil), document.ReasonCodes...),
+			IncompleteValidators: append([]string(nil), document.IncompleteValidators...),
+			ProducerAttemptCount: *document.ProducerAttemptCount, TerminalAction: document.TerminalAction,
+		})
+	}
+	return summaries, nil
+}
+
+func validTerminalAction(value string) bool {
+	switch value {
+	case "accepted", "reject_output", "warn_continue", "fail_run":
+		return true
+	default:
+		return false
+	}
+}
+
+func loadReceipt(path, pipelineID string) (Receipt, error) {
+	var document receiptDocument
+	if err := decodeBoundedJSON(path, maxReceiptBytes, &document); err != nil {
+		return Receipt{}, fmt.Errorf("decode notarius receipt: %w", err)
+	}
+	if document.SchemaVersion != ReceiptSchemaVersion {
+		return Receipt{}, fmt.Errorf("unsupported notarius receipt schema version %q", document.SchemaVersion)
+	}
+	if strings.TrimSpace(document.RunID) == "" || strings.TrimSpace(document.PipelineID) == "" ||
+		strings.TrimSpace(document.OutputDirectory) == "" || strings.TrimSpace(document.ValidationStatus) == "" ||
+		document.NormalizedOutputCount == nil ||
+		document.RejectedOutputCount == nil || document.WarningGroupCount == nil ||
+		document.WarningOccurrenceCount == nil || document.DiagnosticGroupCount == nil ||
+		document.DiagnosticOccurrenceCount == nil || document.DiagnosticsTruncated == nil {
+		return Receipt{}, fmt.Errorf("notarius receipt is missing required fields")
+	}
+	if document.IndexFile != canonicalIndexFile {
+		return Receipt{}, fmt.Errorf("notarius receipt index_file %q is incompatible; want %q", document.IndexFile, canonicalIndexFile)
+	}
+	if document.PipelineID != pipelineID {
+		return Receipt{}, fmt.Errorf("notarius receipt pipeline id %q does not match requested pipeline %q", document.PipelineID, pipelineID)
+	}
+	if *document.NormalizedOutputCount < 0 || *document.RejectedOutputCount < 0 ||
+		*document.WarningGroupCount < 0 || *document.WarningOccurrenceCount < 0 ||
+		*document.DiagnosticGroupCount < 0 || *document.DiagnosticOccurrenceCount < 0 {
+		return Receipt{}, fmt.Errorf("notarius receipt counts must be non-negative")
+	}
+	if !validValidationStatus(document.ValidationStatus) {
+		return Receipt{}, fmt.Errorf("notarius receipt validation_status %q is invalid", document.ValidationStatus)
+	}
+	validationSummaries, err := validateValidationSummaries(document.ValidationSummaries)
+	if err != nil {
+		return Receipt{}, err
+	}
+	if !filepath.IsAbs(document.OutputDirectory) {
+		return Receipt{}, fmt.Errorf("notarius receipt output directory must be absolute")
+	}
+	if document.DebugDirectory != "" && !filepath.IsAbs(document.DebugDirectory) {
+		return Receipt{}, fmt.Errorf("notarius receipt debug directory must be absolute when present")
+	}
+	return Receipt{
+		SchemaVersion:             document.SchemaVersion,
+		RunID:                     document.RunID,
+		PipelineID:                document.PipelineID,
+		OutputDirectory:           filepath.Clean(document.OutputDirectory),
+		IndexFile:                 document.IndexFile,
+		NormalizedOutputCount:     *document.NormalizedOutputCount,
+		RejectedOutputCount:       *document.RejectedOutputCount,
+		WarningGroupCount:         *document.WarningGroupCount,
+		WarningOccurrenceCount:    *document.WarningOccurrenceCount,
+		DiagnosticGroupCount:      *document.DiagnosticGroupCount,
+		DiagnosticOccurrenceCount: *document.DiagnosticOccurrenceCount,
+		DiagnosticsTruncated:      *document.DiagnosticsTruncated,
+		ValidationStatus:          document.ValidationStatus,
+		ValidationSummaries:       validationSummaries,
+		DebugDirectory:            document.DebugDirectory,
+	}, nil
+}
+
+type indexDocument struct {
+	ManifestFile    string            `json:"manifest_file"`
+	OutputFiles     *[]laneDocument   `json:"output_files"`
+	RejectedFile    string            `json:"rejected_file"`
+	WarningsFile    string            `json:"warnings_file"`
+	DiagnosticsFile string            `json:"diagnostics_file"`
+	ChunkMap        *pipelineDocument `json:"chunk_map"`
+	EvidenceContext *pipelineDocument `json:"evidence_context"`
+}
+
+type laneDocument struct {
+	LaneID        string `json:"lane_id"`
+	File          string `json:"file"`
+	MediaType     string `json:"media_type"`
+	ModuleKey     string `json:"module_key"`
+	SchemaID      string `json:"schema_id"`
+	SchemaName    string `json:"schema_name"`
+	SchemaVersion string `json:"schema_version"`
+}
+
+type pipelineDocument struct {
+	ArtifactKind  string `json:"artifact_kind"`
+	File          string `json:"file"`
+	MediaType     string `json:"media_type"`
+	SchemaID      string `json:"schema_id"`
+	SchemaName    string `json:"schema_name"`
+	SchemaVersion string `json:"schema_version"`
+}
+
+func loadIndex(bundleRoot, indexPath string) (Index, error) {
+	var document indexDocument
+	if err := decodeBoundedJSON(indexPath, maxIndexBytes, &document); err != nil {
+		return Index{}, fmt.Errorf("decode notarius index: %w", err)
+	}
+	for _, field := range []struct {
+		name string
+		got  string
+		want string
+	}{
+		{name: "manifest_file", got: document.ManifestFile, want: canonicalManifestFile},
+		{name: "rejected_file", got: document.RejectedFile, want: canonicalRejectedFile},
+		{name: "warnings_file", got: document.WarningsFile, want: canonicalWarningsFile},
+		{name: "diagnostics_file", got: document.DiagnosticsFile, want: canonicalDiagnosticsFile},
+	} {
+		if field.got != field.want {
+			return Index{}, fmt.Errorf("notarius index %s %q is incompatible; want %q", field.name, field.got, field.want)
+		}
+	}
+	if document.OutputFiles == nil {
+		return Index{}, fmt.Errorf("notarius index is missing required output_files")
+	}
+
+	index := Index{
+		Path:            indexPath,
+		ManifestFile:    document.ManifestFile,
+		RejectedFile:    document.RejectedFile,
+		WarningsFile:    document.WarningsFile,
+		DiagnosticsFile: document.DiagnosticsFile,
+	}
+	var err error
+	if index.ManifestPath, err = resolveRegularFile(bundleRoot, index.ManifestFile); err != nil {
+		return Index{}, fmt.Errorf("resolve notarius manifest file: %w", err)
+	}
+	if index.RejectedPath, err = resolveRegularFile(bundleRoot, index.RejectedFile); err != nil {
+		return Index{}, fmt.Errorf("resolve notarius rejection file: %w", err)
+	}
+	if index.WarningsPath, err = resolveRegularFile(bundleRoot, index.WarningsFile); err != nil {
+		return Index{}, fmt.Errorf("resolve notarius warning file: %w", err)
+	}
+	if index.DiagnosticsPath, err = resolveRegularFile(bundleRoot, index.DiagnosticsFile); err != nil {
+		return Index{}, fmt.Errorf("resolve notarius diagnostics file: %w", err)
+	}
+
+	seenLanes := make(map[string]struct{}, len(*document.OutputFiles))
+	for _, lane := range *document.OutputFiles {
+		if strings.TrimSpace(lane.LaneID) == "" || strings.TrimSpace(lane.File) == "" {
+			return Index{}, fmt.Errorf("notarius lane descriptors require lane_id and file")
+		}
+		if _, exists := seenLanes[lane.LaneID]; exists {
+			return Index{}, fmt.Errorf("notarius index contains duplicate lane id %q", lane.LaneID)
+		}
+		seenLanes[lane.LaneID] = struct{}{}
+		path, err := resolveRegularFile(bundleRoot, lane.File)
+		if err != nil {
+			return Index{}, fmt.Errorf("resolve notarius lane %q file: %w", lane.LaneID, err)
+		}
+		index.Lanes = append(index.Lanes, LaneDescriptor{
+			LaneID: lane.LaneID, File: lane.File, Path: path, MediaType: lane.MediaType,
+			ModuleKey: lane.ModuleKey, SchemaID: lane.SchemaID, SchemaName: lane.SchemaName,
+			SchemaVersion: lane.SchemaVersion,
+		})
+	}
+	if document.ChunkMap != nil {
+		index.ChunkMap, err = resolvePipelineDescriptor(bundleRoot, "chunk_map", *document.ChunkMap)
+		if err != nil {
+			return Index{}, err
+		}
+	}
+	if document.EvidenceContext != nil {
+		index.EvidenceContext, err = resolvePipelineDescriptor(bundleRoot, "evidence_context", *document.EvidenceContext)
+		if err != nil {
+			return Index{}, err
+		}
+	}
+	return index, nil
+}
+
+func resolvePipelineDescriptor(bundleRoot, label string, document pipelineDocument) (*PipelineDescriptor, error) {
+	if strings.TrimSpace(document.ArtifactKind) == "" || strings.TrimSpace(document.File) == "" ||
+		strings.TrimSpace(document.MediaType) == "" || strings.TrimSpace(document.SchemaID) == "" ||
+		strings.TrimSpace(document.SchemaName) == "" || strings.TrimSpace(document.SchemaVersion) == "" {
+		return nil, fmt.Errorf("notarius %s descriptor is missing required fields", label)
+	}
+	path, err := resolveRegularFile(bundleRoot, document.File)
+	if err != nil {
+		return nil, fmt.Errorf("resolve notarius %s file: %w", label, err)
+	}
+	return &PipelineDescriptor{
+		ArtifactKind: document.ArtifactKind, File: document.File, Path: path,
+		MediaType: document.MediaType, SchemaID: document.SchemaID,
+		SchemaName: document.SchemaName, SchemaVersion: document.SchemaVersion,
+	}, nil
+}
+
+type rejectionDocument struct {
+	Rejected *[]struct {
+		Stage         string `json:"stage"`
+		StepID        string `json:"step_id"`
+		LaneID        string `json:"lane_id"`
+		ModuleKey     string `json:"module_key"`
+		ChunkID       string `json:"chunk_id"`
+		ValidatorName string `json:"validator_name"`
+		ReasonCode    string `json:"reason_code"`
+		Message       string `json:"message"`
+	} `json:"rejected"`
+}
+
+func loadRejections(path string) ([]RejectionSummary, error) {
+	var document rejectionDocument
+	if err := decodeBoundedJSON(path, maxSummaryBytes, &document); err != nil {
+		return nil, fmt.Errorf("decode notarius rejections: %w", err)
+	}
+	if document.Rejected == nil {
+		return nil, fmt.Errorf("notarius rejection document is missing rejected array")
+	}
+	summaries := make([]RejectionSummary, 0, len(*document.Rejected))
+	for _, item := range *document.Rejected {
+		if strings.TrimSpace(item.Stage) == "" || strings.TrimSpace(item.Message) == "" {
+			return nil, fmt.Errorf("notarius rejection entries require stage and message")
+		}
+		summaries = append(summaries, RejectionSummary{
+			Stage: item.Stage, StepID: item.StepID, LaneID: item.LaneID,
+			ModuleKey: item.ModuleKey, ChunkID: item.ChunkID,
+			ValidatorName: item.ValidatorName, ReasonCode: item.ReasonCode,
+		})
+	}
+	return summaries, nil
+}
+
+type findingGroupDocument struct {
+	Disposition     string                   `json:"disposition"`
+	Category        string                   `json:"category"`
+	ReasonCode      string                   `json:"reason_code"`
+	Origin          diagnosticOriginDocument `json:"origin"`
+	OccurrenceCount *int                     `json:"occurrence_count"`
+	Samples         *[]struct {
+		Scope      string `json:"scope"`
+		Message    string `json:"message"`
+		ChunkID    string `json:"chunk_id"`
+		ChunkIndex *int   `json:"chunk_index"`
+	} `json:"samples"`
+	OmittedSampleCount *int `json:"omitted_sample_count"`
+}
+
+type diagnosticOriginDocument struct {
+	Stage        string `json:"stage"`
+	StepID       string `json:"step_id"`
+	LaneID       string `json:"lane_id"`
+	ModuleKey    string `json:"module_key"`
+	ValidatorKey string `json:"validator_key"`
+}
+
+type warningDocument struct {
+	SchemaVersion   string                  `json:"schema_version"`
+	GroupCount      *int                    `json:"group_count"`
+	OccurrenceCount *int                    `json:"occurrence_count"`
+	Groups          *[]findingGroupDocument `json:"groups"`
+}
+
+func loadWarnings(path string) ([]WarningSummary, error) {
+	var document warningDocument
+	if err := decodeBoundedJSON(path, maxSummaryBytes, &document); err != nil {
+		return nil, fmt.Errorf("decode notarius warnings: %w", err)
+	}
+	if document.SchemaVersion != warningsSchemaVersion || document.GroupCount == nil ||
+		document.OccurrenceCount == nil || document.Groups == nil {
+		return nil, fmt.Errorf("notarius warning document is missing or incompatible required fields")
+	}
+	if *document.GroupCount < 0 || *document.GroupCount > maxWarningGroups || *document.OccurrenceCount < 0 ||
+		*document.GroupCount != len(*document.Groups) {
+		return nil, fmt.Errorf("notarius warning document counts are inconsistent")
+	}
+	summaries := make([]WarningSummary, 0, len(*document.Groups))
+	occurrences := 0
+	for _, group := range *document.Groups {
+		if err := validateFindingGroup(group); err != nil {
+			return nil, fmt.Errorf("notarius warning group: %w", err)
+		}
+		if group.Disposition != "warning" {
+			return nil, fmt.Errorf("notarius warning group disposition %q is invalid", group.Disposition)
+		}
+		if *group.OccurrenceCount > int(^uint(0)>>1)-occurrences {
+			return nil, fmt.Errorf("notarius warning occurrence count overflows")
+		}
+		occurrences += *group.OccurrenceCount
+		summaries = append(summaries, WarningSummary{
+			Disposition: group.Disposition, Category: group.Category, ReasonCode: group.ReasonCode,
+			Origin: diagnosticOrigin(group.Origin), OccurrenceCount: *group.OccurrenceCount,
+		})
+	}
+	if occurrences != *document.OccurrenceCount {
+		return nil, fmt.Errorf("notarius warning document occurrence count is inconsistent")
+	}
+	return summaries, nil
+}
+
+type diagnosticDocument struct {
+	SchemaVersion                string                  `json:"schema_version"`
+	GroupCount                   *int                    `json:"group_count"`
+	OccurrenceCount              *int                    `json:"occurrence_count"`
+	Truncated                    *bool                   `json:"truncated"`
+	UnrepresentedOccurrenceCount *int                    `json:"unrepresented_occurrence_count"`
+	Groups                       *[]findingGroupDocument `json:"groups"`
+}
+
+func loadDiagnostics(path string) ([]DiagnosticSummary, int, bool, error) {
+	var document diagnosticDocument
+	if err := decodeBoundedJSON(path, maxSummaryBytes, &document); err != nil {
+		return nil, 0, false, fmt.Errorf("decode notarius diagnostics: %w", err)
+	}
+	if document.SchemaVersion != diagnosticsSchemaVersion || document.GroupCount == nil ||
+		document.OccurrenceCount == nil || document.Truncated == nil ||
+		document.UnrepresentedOccurrenceCount == nil || document.Groups == nil {
+		return nil, 0, false, fmt.Errorf("notarius diagnostics document is missing or incompatible required fields")
+	}
+	if *document.GroupCount < 0 || *document.GroupCount > maxDiagnosticGroups || *document.OccurrenceCount < 0 ||
+		*document.UnrepresentedOccurrenceCount < 0 || *document.GroupCount != len(*document.Groups) {
+		return nil, 0, false, fmt.Errorf("notarius diagnostics document counts are inconsistent")
+	}
+	if !*document.Truncated && *document.UnrepresentedOccurrenceCount != 0 {
+		return nil, 0, false, fmt.Errorf("notarius diagnostics document has unrepresented occurrences without truncation")
+	}
+	summaries := make([]DiagnosticSummary, 0, len(*document.Groups))
+	representedOccurrences := 0
+	for _, group := range *document.Groups {
+		if err := validateFindingGroup(group); err != nil {
+			return nil, 0, false, fmt.Errorf("notarius diagnostic group: %w", err)
+		}
+		if group.Disposition != "advisory" && group.Disposition != "observation" {
+			return nil, 0, false, fmt.Errorf("notarius diagnostic group disposition %q is invalid", group.Disposition)
+		}
+		if *group.OccurrenceCount > int(^uint(0)>>1)-representedOccurrences {
+			return nil, 0, false, fmt.Errorf("notarius diagnostic occurrence count overflows")
+		}
+		representedOccurrences += *group.OccurrenceCount
+		summaries = append(summaries, DiagnosticSummary{
+			Disposition: group.Disposition, Category: group.Category, ReasonCode: group.ReasonCode,
+			Origin: diagnosticOrigin(group.Origin), OccurrenceCount: *group.OccurrenceCount,
+		})
+	}
+	if *document.UnrepresentedOccurrenceCount > int(^uint(0)>>1)-representedOccurrences ||
+		representedOccurrences+*document.UnrepresentedOccurrenceCount != *document.OccurrenceCount {
+		return nil, 0, false, fmt.Errorf("notarius diagnostics document occurrence count is inconsistent")
+	}
+	return summaries, *document.OccurrenceCount, *document.Truncated, nil
+}
+
+func validateFindingGroup(group findingGroupDocument) error {
+	if strings.TrimSpace(group.Disposition) == "" || strings.TrimSpace(group.Category) == "" ||
+		strings.TrimSpace(group.ReasonCode) == "" || !validDiagnosticOriginStage(group.Origin.Stage) ||
+		!validDiagnosticCategory(group.Disposition, group.Category) ||
+		group.OccurrenceCount == nil || *group.OccurrenceCount <= 0 || group.Samples == nil ||
+		group.OmittedSampleCount == nil || *group.OmittedSampleCount < 0 {
+		return fmt.Errorf("missing required fields")
+	}
+	if len(*group.Samples) == 0 || len(*group.Samples) > maxFindingSamples ||
+		*group.OmittedSampleCount != *group.OccurrenceCount-len(*group.Samples) {
+		return fmt.Errorf("sample counts are inconsistent")
+	}
+	for _, sample := range *group.Samples {
+		if strings.TrimSpace(sample.Scope) == "" || strings.TrimSpace(sample.Message) == "" ||
+			(sample.ChunkIndex != nil && *sample.ChunkIndex < 0) {
+			return fmt.Errorf("samples require scope and message")
+		}
+	}
+	return nil
+}
+
+func validDiagnosticCategory(disposition, category string) bool {
+	switch disposition {
+	case "warning":
+		return category == "configuration" || category == "degradation" ||
+			category == "validation_incomplete" || category == "fallback"
+	case "advisory":
+		return category == "data_quality"
+	case "observation":
+		return category == "normalization"
+	default:
+		return false
+	}
+}
+
+func validDiagnosticOriginStage(stage string) bool {
+	switch stage {
+	case "references", "chunk", "extract", "merge", "normalize":
+		return true
+	default:
+		return false
+	}
+}
+
+func diagnosticOrigin(document diagnosticOriginDocument) DiagnosticOrigin {
+	return DiagnosticOrigin{
+		Stage: document.Stage, StepID: document.StepID, LaneID: document.LaneID,
+		ModuleKey: document.ModuleKey, ValidatorKey: document.ValidatorKey,
+	}
+}
+
+func sumWarningOccurrences(values []WarningSummary) (int, error) {
+	total := 0
+	for _, value := range values {
+		if value.OccurrenceCount > int(^uint(0)>>1)-total {
+			return 0, fmt.Errorf("notarius warning occurrence count overflows")
+		}
+		total += value.OccurrenceCount
+	}
+	return total, nil
+}
+
+func decodeBoundedJSON(path string, limit int64, destination any) error {
+	data, err := fileops.ReadRegularFile(path, limit)
+	if err != nil {
+		return fmt.Errorf("notarius JSON result exceeds or cannot be read within %d-byte limit: %w", limit, err)
+	}
+	if err := json.Unmarshal(data, destination); err != nil {
+		return err
+	}
+	return nil
+}
+
+func validateBundleRoot(outputRoot, bundleRoot string) (string, error) {
+	root := filepath.Clean(outputRoot)
+	bundle := filepath.Clean(bundleRoot)
+	relative, err := filepath.Rel(root, bundle)
+	if err != nil {
+		return "", fmt.Errorf("compare notarius output paths: %w", err)
+	}
+	if relative == "." || relative == ".." || strings.HasPrefix(relative, ".."+string(filepath.Separator)) {
+		return "", fmt.Errorf("notarius output directory %q is not beneath output root %q", bundleRoot, outputRoot)
+	}
+	if err := requireDirectoryTree(root, relative); err != nil {
+		return "", fmt.Errorf("validate notarius output directory: %w", err)
+	}
+	return bundle, nil
+}
+
+func resolveRegularFile(root, logicalPath string) (string, error) {
+	resolved, err := pathsafe.JoinSlashRelativeUnderRoot(root, logicalPath)
+	if err != nil {
+		return "", err
+	}
+	relative, err := filepath.Rel(root, resolved)
+	if err != nil {
+		return "", err
+	}
+	if err := requireRegularFileTree(root, relative); err != nil {
+		return "", err
+	}
+	return resolved, nil
+}
+
+func requireDirectoryTree(root, relative string) error {
+	if err := requireDirectory(root); err != nil {
+		return err
+	}
+	current := root
+	for _, component := range strings.Split(relative, string(filepath.Separator)) {
+		current = filepath.Join(current, component)
+		if err := requireDirectory(current); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func requireRegularFileTree(root, relative string) error {
+	components := strings.Split(relative, string(filepath.Separator))
+	if len(components) == 0 {
+		return fmt.Errorf("regular file path is required")
+	}
+	if err := requireDirectory(root); err != nil {
+		return err
+	}
+	current := root
+	for _, component := range components[:len(components)-1] {
+		current = filepath.Join(current, component)
+		if err := requireDirectory(current); err != nil {
+			return err
+		}
+	}
+	return requireRegularFile(filepath.Join(current, components[len(components)-1]))
+}
+
+func requireDirectory(path string) error {
+	info, err := os.Lstat(path)
+	if err != nil {
+		return err
+	}
+	if info.Mode()&os.ModeSymlink != 0 || !info.IsDir() {
+		return fmt.Errorf("path %q must be a directory without symlinks", path)
+	}
+	return nil
+}
+
+func requireRegularFile(path string) error {
+	info, err := os.Lstat(path)
+	if err != nil {
+		return err
+	}
+	if info.Mode()&os.ModeSymlink != 0 || !info.Mode().IsRegular() {
+		return fmt.Errorf("path %q must be a regular file without symlinks", path)
+	}
+	return nil
+}
+
+func validateLogDestination(path string) error {
+	if err := requireDirectory(filepath.Dir(path)); err != nil {
+		return err
+	}
+	info, err := os.Lstat(path)
+	if errors.Is(err, os.ErrNotExist) {
+		return nil
+	}
+	if err != nil {
+		return err
+	}
+	if info.Mode()&os.ModeSymlink != 0 || !info.Mode().IsRegular() {
+		return fmt.Errorf("path %q must be absent or a regular file without symlinks", path)
+	}
+	return nil
+}

internal/adapters/notarius/subprocess_test.go

@@ -0,0 +1,716 @@
+package notarius
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"os"
+	"path/filepath"
+	"reflect"
+	"strings"
+	"testing"
+	"time"
+
+	sharedsubprocess "gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+)
+
+func TestSubprocessRunnerBuildsExactInvocationAndDiscoversBundle(t *testing.T) {
+	req := validRunRequest(t)
+	var captured sharedsubprocess.RunRequest
+	runner := &SubprocessRunner{run: func(_ context.Context, processReq sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+		captured = processReq
+		writeValidBundleAndReceipt(t, req, true)
+		return sharedsubprocess.RunResult{ExitCode: 0, Duration: 2 * time.Second}, nil
+	}}
+
+	result, err := runner.Run(context.Background(), req)
+	if err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	wantArgs := []string{
+		"run", "dnd-session", "--config", req.ConfigPath, "--input", req.InputPath,
+		"--output-dir", req.OutputRoot, "--json",
+	}
+	if !reflect.DeepEqual(captured.Args, wantArgs) {
+		t.Fatalf("subprocess args = %#v, want %#v", captured.Args, wantArgs)
+	}
+	if captured.Executable != req.Binary || captured.WorkingDir != req.WorkingDirectory || captured.Timeout != req.Timeout {
+		t.Fatalf("subprocess request = %#v", captured)
+	}
+	if captured.StdoutLogPath != req.ReceiptPath || captured.StderrLogPath != req.LogPath {
+		t.Fatalf("stream paths = stdout %q stderr %q", captured.StdoutLogPath, captured.StderrLogPath)
+	}
+	if captured.EnvOverrides != nil {
+		t.Fatalf("environment overrides = %#v, want inherited environment only", captured.EnvOverrides)
+	}
+	for _, arg := range captured.Args {
+		if arg == "--session-id" {
+			t.Fatal("subprocess args unexpectedly contain --session-id")
+		}
+	}
+
+	if result.Receipt.SchemaVersion != ReceiptSchemaVersion || result.Receipt.RunID != "notarius-run-1" {
+		t.Fatalf("receipt = %#v", result.Receipt)
+	}
+	if len(result.Receipt.ValidationSummaries) != 1 || result.Receipt.ValidationSummaries[0].LaneID != "npc-registry" ||
+		result.Receipt.ValidationSummaries[0].Status != "complete" {
+		t.Fatalf("validation summaries = %#v", result.Receipt.ValidationSummaries)
+	}
+	if len(result.Index.Lanes) != 1 || result.Index.Lanes[0].LaneID != "npc-registry" {
+		t.Fatalf("lanes = %#v", result.Index.Lanes)
+	}
+	if result.Index.ChunkMap == nil || result.Index.ChunkMap.ArtifactKind != "chunk_map" {
+		t.Fatalf("chunk map = %#v", result.Index.ChunkMap)
+	}
+	if result.Index.EvidenceContext == nil || result.Index.EvidenceContext.ArtifactKind != "evidence_context" {
+		t.Fatalf("evidence context = %#v", result.Index.EvidenceContext)
+	}
+	if len(result.Rejections) != 1 || result.Rejections[0].LaneID != "spells" || result.Rejections[0].ReasonCode != "invalid_spell" {
+		t.Fatalf("rejections = %#v", result.Rejections)
+	}
+	if len(result.Warnings) != 1 || result.Warnings[0].Category != "degradation" || result.Warnings[0].ReasonCode != "normalized_name" {
+		t.Fatalf("warnings = %#v", result.Warnings)
+	}
+	if len(result.Diagnostics) != 1 || result.Diagnostics[0].Category != "data_quality" || result.Diagnostics[0].ReasonCode != "low_confidence" {
+		t.Fatalf("diagnostics = %#v", result.Diagnostics)
+	}
+}
+
+func TestSubprocessRunnerBuildsOrderedReferenceArguments(t *testing.T) {
+	req := validRunRequest(t)
+	referenceRoot := t.TempDir()
+	req.References = []ReferenceBinding{
+		{Selector: " party ", Path: filepath.Join(referenceRoot, "party context=primary.json")},
+		{Selector: " npc-registry . extract . glossary ", Path: filepath.Join(referenceRoot, "glossary.json")},
+	}
+	originalReferences := append([]ReferenceBinding(nil), req.References...)
+	var captured sharedsubprocess.RunRequest
+	runner := &SubprocessRunner{run: func(_ context.Context, processReq sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+		captured = processReq
+		writeValidBundleAndReceipt(t, req, false)
+		return sharedsubprocess.RunResult{ExitCode: 0}, nil
+	}}
+
+	if _, err := runner.Run(context.Background(), req); err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	wantArgs := []string{
+		"run", req.PipelineID,
+		"--config", req.ConfigPath,
+		"--input", req.InputPath,
+		"--output-dir", req.OutputRoot,
+		"--reference", "party=" + req.References[0].Path,
+		"--reference", "npc-registry.extract.glossary=" + req.References[1].Path,
+		"--json",
+	}
+	if !reflect.DeepEqual(captured.Args, wantArgs) {
+		t.Fatalf("subprocess args = %#v, want %#v", captured.Args, wantArgs)
+	}
+	if !reflect.DeepEqual(req.References, originalReferences) {
+		t.Fatalf("Run() mutated caller references = %#v, want %#v", req.References, originalReferences)
+	}
+}
+
+func TestSubprocessRunnerRejectsInvalidReferencesBeforeLaunch(t *testing.T) {
+	tests := []struct {
+		name       string
+		references func(string) []ReferenceBinding
+		wantErr    string
+	}{
+		{
+			name: "invalid selector",
+			references: func(root string) []ReferenceBinding {
+				return []ReferenceBinding{{Selector: "lane.prepare.party", Path: filepath.Join(root, "party.json")}}
+			},
+			wantErr: "selector",
+		},
+		{
+			name: "duplicate normalized selector",
+			references: func(root string) []ReferenceBinding {
+				return []ReferenceBinding{
+					{Selector: "lane.party", Path: filepath.Join(root, "party.json")},
+					{Selector: " lane . party ", Path: filepath.Join(root, "party-2.json")},
+				}
+			},
+			wantErr: "duplicated",
+		},
+		{
+			name: "empty path",
+			references: func(string) []ReferenceBinding {
+				return []ReferenceBinding{{Selector: "party", Path: "  "}}
+			},
+			wantErr: "path is required",
+		},
+		{
+			name: "relative path",
+			references: func(string) []ReferenceBinding {
+				return []ReferenceBinding{{Selector: "party", Path: "references/party.json"}}
+			},
+			wantErr: "path must be absolute",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			req := validRunRequest(t)
+			req.References = tt.references(t.TempDir())
+			started := false
+			runner := &SubprocessRunner{run: func(context.Context, sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+				started = true
+				return sharedsubprocess.RunResult{}, nil
+			}}
+
+			_, err := runner.Run(context.Background(), req)
+			if err == nil || !strings.Contains(err.Error(), tt.wantErr) {
+				t.Fatalf("Run() error = %v, want containing %q", err, tt.wantErr)
+			}
+			if started {
+				t.Fatal("subprocess started after request validation failure")
+			}
+		})
+	}
+}
+
+func TestSubprocessRunnerUsesMinimalEnvironmentAndSeparatesStreams(t *testing.T) {
+	req := validRunRequest(t)
+	writeValidBundleAndReceipt(t, req, false)
+	receiptFixture := req.ReceiptPath + ".fixture"
+	data, err := os.ReadFile(req.ReceiptPath)
+	if err != nil {
+		t.Fatalf("ReadFile(receipt) error = %v", err)
+	}
+	if err := os.WriteFile(receiptFixture, data, 0o644); err != nil {
+		t.Fatalf("WriteFile(receipt fixture) error = %v", err)
+	}
+	if err := os.Remove(req.ReceiptPath); err != nil {
+		t.Fatalf("Remove(receipt) error = %v", err)
+	}
+
+	captureDir := filepath.Join(filepath.Dir(req.ReceiptPath), "capture")
+	if err := os.Mkdir(captureDir, 0o755); err != nil {
+		t.Fatalf("Mkdir(capture) error = %v", err)
+	}
+	script := writeShellScript(t, `#!/bin/sh
+pwd > "$NOTARIUS_CAPTURE_DIR/working-directory"
+printf '%s' "$NOTARIUS_INHERITED_VALUE" > "$NOTARIUS_CAPTURE_DIR/environment"
+printf 'diagnostic stream\n' >&2
+cat "$NOTARIUS_RECEIPT_FIXTURE"
+`)
+	req.Binary = script
+	t.Setenv("NOTARIUS_CAPTURE_DIR", captureDir)
+	t.Setenv("NOTARIUS_INHERITED_VALUE", "inherited-value")
+	t.Setenv("NOTARIUS_RECEIPT_FIXTURE", receiptFixture)
+
+	if _, err := NewSubprocessRunner().Run(context.Background(), req); err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+	assertTextFile(t, filepath.Join(captureDir, "working-directory"), req.WorkingDirectory+"\n")
+	assertTextFile(t, filepath.Join(captureDir, "environment"), "")
+	assertTextFile(t, req.LogPath, "diagnostic stream\n")
+	receiptBytes, err := os.ReadFile(req.ReceiptPath)
+	if err != nil {
+		t.Fatalf("ReadFile(receipt) error = %v", err)
+	}
+	if strings.Contains(string(receiptBytes), "diagnostic stream") {
+		t.Fatal("receipt contains stderr output")
+	}
+}
+
+func TestSubprocessRunnerReturnsProcessFailuresWithoutParsingStdout(t *testing.T) {
+	tests := []struct {
+		name       string
+		scriptBody string
+		timeout    time.Duration
+		cancel     bool
+		want       string
+	}{
+		{name: "nonzero", scriptBody: "printf '{malformed receipt'; printf 'failed\\n' >&2; exit 7\n", timeout: time.Second, want: "exit code 7"},
+		{name: "timeout", scriptBody: "sleep 5\n", timeout: 20 * time.Millisecond, want: "timed out"},
+		{name: "cancellation", scriptBody: "sleep 5\n", timeout: time.Second, cancel: true, want: "canceled"},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			req := validRunRequest(t)
+			req.Binary = writeShellScript(t, "#!/bin/sh\n"+test.scriptBody)
+			req.Timeout = test.timeout
+			ctx := context.Background()
+			if test.cancel {
+				cancelCtx, cancel := context.WithCancel(ctx)
+				ctx = cancelCtx
+				time.AfterFunc(20*time.Millisecond, cancel)
+			}
+			_, err := NewSubprocessRunner().Run(ctx, req)
+			if err == nil || !strings.Contains(err.Error(), test.want) {
+				t.Fatalf("Run() error = %v, want fragment %q", err, test.want)
+			}
+			if strings.Contains(err.Error(), "decode notarius receipt") {
+				t.Fatalf("Run() parsed stdout after process failure: %v", err)
+			}
+		})
+	}
+}
+
+func TestSubprocessRunnerReturnsSharedSubprocessErrorWithoutReadingReceipt(t *testing.T) {
+	req := validRunRequest(t)
+	if err := os.WriteFile(req.ReceiptPath, []byte("not json"), 0o644); err != nil {
+		t.Fatalf("WriteFile(receipt) error = %v", err)
+	}
+	wantErr := errors.New("process failed")
+	runner := &SubprocessRunner{run: func(context.Context, sharedsubprocess.RunRequest) (sharedsubprocess.RunResult, error) {
+		return sharedsubprocess.RunResult{ExitCode: 9}, wantErr
+	}}
+	_, err := runner.Run(context.Background(), req)
+	if !errors.Is(err, wantErr) {
+		t.Fatalf("Run() error = %v, want wrapped process error", err)
+	}
+	if strings.Contains(err.Error(), "decode") {
+		t.Fatalf("Run() parsed receipt after failure: %v", err)
+	}
+}
+
+func TestLoadReceiptValidation(t *testing.T) {
+	root := t.TempDir()
+	valid := map[string]any{
+		"schema_version": ReceiptSchemaVersion, "run_id": "run-1", "pipeline_id": "pipeline-1",
+		"output_directory": filepath.Join(root, "outputs", "run-1"), "index_file": "index.json",
+		"normalized_output_count": 1, "rejected_output_count": 0,
+		"warning_group_count": 0, "warning_occurrence_count": 0,
+		"diagnostic_group_count": 0, "diagnostic_occurrence_count": 0,
+		"diagnostics_truncated": false, "validation_status": "approved", "future_field": true,
+	}
+	tests := []struct {
+		name      string
+		mutate    func(map[string]any)
+		raw       []byte
+		wantOK    bool
+		wantError string
+	}{
+		{name: "unknown fields tolerated", wantOK: true},
+		{name: "malformed", raw: []byte("{")},
+		{name: "unsupported version", mutate: func(v map[string]any) { v["schema_version"] = "notarius.run-result.v1" }},
+		{name: "missing field", mutate: func(v map[string]any) { delete(v, "run_id") }},
+		{name: "pipeline mismatch", mutate: func(v map[string]any) { v["pipeline_id"] = "other" }},
+		{name: "relative output", mutate: func(v map[string]any) { v["output_directory"] = "run-1" }},
+		{name: "negative count", mutate: func(v map[string]any) { v["warning_group_count"] = -1 }},
+		{name: "invalid validation status", mutate: func(v map[string]any) { v["validation_status"] = "valid" }},
+		{
+			name: "nested index", mutate: func(v map[string]any) { v["index_file"] = "nested/index.json" },
+			wantError: `index_file "nested/index.json"`,
+		},
+		{
+			name: "cleanable index", mutate: func(v map[string]any) { v["index_file"] = "./index.json" },
+			wantError: `index_file "./index.json"`,
+		},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			path := filepath.Join(root, strings.ReplaceAll(test.name, " ", "-")+".json")
+			values := cloneMap(valid)
+			if test.mutate != nil {
+				test.mutate(values)
+			}
+			if test.raw != nil {
+				if err := os.WriteFile(path, test.raw, 0o644); err != nil {
+					t.Fatalf("WriteFile() error = %v", err)
+				}
+			} else {
+				writeJSONFile(t, path, values)
+			}
+			_, err := loadReceipt(path, "pipeline-1")
+			if test.wantOK && err != nil {
+				t.Fatalf("loadReceipt() error = %v", err)
+			}
+			if !test.wantOK && err == nil {
+				t.Fatal("loadReceipt() error = nil, want validation failure")
+			}
+			if test.wantError != "" && !strings.Contains(err.Error(), test.wantError) {
+				t.Fatalf("loadReceipt() error = %v, want fragment %q", err, test.wantError)
+			}
+		})
+	}
+
+	oversized := filepath.Join(root, "oversized.json")
+	if err := os.WriteFile(oversized, []byte(strings.Repeat("x", maxReceiptBytes+1)), 0o644); err != nil {
+		t.Fatalf("WriteFile(oversized) error = %v", err)
+	}
+	if _, err := loadReceipt(oversized, "pipeline-1"); err == nil || !strings.Contains(err.Error(), "exceeds") {
+		t.Fatalf("loadReceipt(oversized) error = %v", err)
+	}
+}
+
+func TestValidateBundleRootRejectsEscapesAndSymlinks(t *testing.T) {
+	root := t.TempDir()
+	outputRoot := filepath.Join(root, "output")
+	if err := os.Mkdir(outputRoot, 0o755); err != nil {
+		t.Fatalf("Mkdir(output root) error = %v", err)
+	}
+	validBundle := filepath.Join(outputRoot, "run-1")
+	if err := os.Mkdir(validBundle, 0o755); err != nil {
+		t.Fatalf("Mkdir(bundle) error = %v", err)
+	}
+	if _, err := validateBundleRoot(outputRoot, validBundle); err != nil {
+		t.Fatalf("validateBundleRoot(valid) error = %v", err)
+	}
+
+	outside := filepath.Join(root, "output-other")
+	if err := os.Mkdir(outside, 0o755); err != nil {
+		t.Fatalf("Mkdir(outside) error = %v", err)
+	}
+	for name, candidate := range map[string]string{"equal root": outputRoot, "escape": root, "prefix confusion": outside} {
+		t.Run(name, func(t *testing.T) {
+			if _, err := validateBundleRoot(outputRoot, candidate); err == nil {
+				t.Fatalf("validateBundleRoot(%q) error = nil", candidate)
+			}
+		})
+	}
+	symlink := filepath.Join(outputRoot, "linked")
+	if err := os.Symlink(outside, symlink); err != nil {
+		t.Skipf("Symlink() unavailable: %v", err)
+	}
+	if _, err := validateBundleRoot(outputRoot, symlink); err == nil {
+		t.Fatal("validateBundleRoot(symlink) error = nil")
+	}
+}
+
+func TestLoadIndexRejectsMalformedUnsafeAndUnsupportedDocuments(t *testing.T) {
+	tests := []struct {
+		name       string
+		indexValue any
+		prepare    func(*testing.T, string)
+		wantError  string
+	}{
+		{name: "malformed", indexValue: json.RawMessage(`{"manifest_file":`)},
+		{name: "unsupported output shape", indexValue: map[string]any{"manifest_file": "manifest.json", "output_files": map[string]any{}, "rejected_file": "rejected.json", "warnings_file": "warnings.json"}},
+		{name: "missing management path", indexValue: map[string]any{"output_files": []any{}, "rejected_file": "rejected.json", "warnings_file": "warnings.json"}},
+		{name: "renamed manifest", indexValue: func() any {
+			value := validIndexValue([]any{})
+			value["manifest_file"] = "metadata.json"
+			return value
+		}(), wantError: `manifest_file "metadata.json"`},
+		{name: "cleanable manifest", indexValue: func() any {
+			value := validIndexValue([]any{})
+			value["manifest_file"] = "./manifest.json"
+			return value
+		}(), wantError: `manifest_file "./manifest.json"`},
+		{name: "renamed rejections", indexValue: func() any {
+			value := validIndexValue([]any{})
+			value["rejected_file"] = "rejections.json"
+			return value
+		}(), wantError: `rejected_file "rejections.json"`},
+		{name: "renamed warnings", indexValue: func() any {
+			value := validIndexValue([]any{})
+			value["warnings_file"] = "diagnostics/warnings.json"
+			return value
+		}(), wantError: `warnings_file "diagnostics/warnings.json"`},
+		{name: "duplicate lane", indexValue: validIndexValue([]any{
+			map[string]any{"lane_id": "npc", "file": "lanes/npc.json"},
+			map[string]any{"lane_id": "npc", "file": "lanes/npc.json"},
+		})},
+		{name: "absolute logical path", indexValue: validIndexValue([]any{map[string]any{"lane_id": "npc", "file": "/tmp/npc.json"}})},
+		{name: "lexical traversal", indexValue: validIndexValue([]any{map[string]any{"lane_id": "npc", "file": "../outside.json"}})},
+		{name: "root prefix confusion", indexValue: validIndexValue([]any{map[string]any{"lane_id": "npc", "file": "../bundle-other/npc.json"}})},
+		{name: "file symlink", indexValue: validIndexValue([]any{map[string]any{"lane_id": "npc", "file": "lanes/npc.json"}}), prepare: func(t *testing.T, bundle string) {
+			if err := os.Symlink(filepath.Join(bundle, "manifest.json"), filepath.Join(bundle, "lanes", "npc.json")); err != nil {
+				t.Skipf("Symlink() unavailable: %v", err)
+			}
+		}},
+		{name: "directory symlink", indexValue: validIndexValue([]any{map[string]any{"lane_id": "npc", "file": "linked/npc.json"}}), prepare: func(t *testing.T, bundle string) {
+			if err := os.Symlink(filepath.Join(bundle, "lanes"), filepath.Join(bundle, "linked")); err != nil {
+				t.Skipf("Symlink() unavailable: %v", err)
+			}
+		}},
+		{name: "missing management file", indexValue: validIndexValue([]any{}), prepare: func(t *testing.T, bundle string) {
+			if err := os.Remove(filepath.Join(bundle, "manifest.json")); err != nil {
+				t.Fatalf("Remove(manifest) error = %v", err)
+			}
+		}},
+		{name: "incomplete pipeline descriptor", indexValue: func() any {
+			value := validIndexValue([]any{})
+			value["chunk_map"] = map[string]any{"artifact_kind": "chunk_map", "file": "chunk-map.json"}
+			return value
+		}()},
+		{name: "pipeline descriptor escape", indexValue: func() any {
+			value := validIndexValue([]any{})
+			value["evidence_context"] = map[string]any{
+				"artifact_kind": "evidence_context", "file": "../evidence.json", "media_type": "application/json",
+				"schema_id": "evidence", "schema_name": "Evidence", "schema_version": "v1",
+			}
+			return value
+		}()},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			bundle := createBundleSkeleton(t)
+			indexPath := filepath.Join(bundle, "index.json")
+			if raw, ok := test.indexValue.(json.RawMessage); ok {
+				if err := os.WriteFile(indexPath, raw, 0o644); err != nil {
+					t.Fatalf("WriteFile(index) error = %v", err)
+				}
+			} else {
+				writeJSONFile(t, indexPath, test.indexValue)
+			}
+			if test.prepare != nil {
+				test.prepare(t, bundle)
+			}
+			if _, err := loadIndex(bundle, indexPath); err == nil {
+				t.Fatal("loadIndex() error = nil, want failure")
+			} else if test.wantError != "" && !strings.Contains(err.Error(), test.wantError) {
+				t.Fatalf("loadIndex() error = %v, want fragment %q", err, test.wantError)
+			}
+		})
+	}
+
+	bundle := createBundleSkeleton(t)
+	oversizedIndex := filepath.Join(bundle, "index.json")
+	if err := os.WriteFile(oversizedIndex, []byte(strings.Repeat("x", maxIndexBytes+1)), 0o644); err != nil {
+		t.Fatalf("WriteFile(oversized index) error = %v", err)
+	}
+	if _, err := loadIndex(bundle, oversizedIndex); err == nil || !strings.Contains(err.Error(), "exceeds") {
+		t.Fatalf("loadIndex(oversized) error = %v", err)
+	}
+}
+
+func TestLoadDiagnosticSummariesValidateBoundsAndTolerateUnknownFields(t *testing.T) {
+	root := t.TempDir()
+	rejectedPath := filepath.Join(root, "rejected.json")
+	warningsPath := filepath.Join(root, "warnings.json")
+	diagnosticsPath := filepath.Join(root, "diagnostics.json")
+	writeJSONFile(t, rejectedPath, map[string]any{"rejected": []any{map[string]any{
+		"stage": "validate", "lane_id": "spells", "reason_code": "invalid", "message": "do not retain this", "future": true,
+	}}, "future": true})
+	writeJSONFile(t, warningsPath, findingEnvelope(warningsSchemaVersion, []any{findingGroup("warning", "degradation", "bounded", "normalize", 2)}, 2, false, 0))
+	writeJSONFile(t, diagnosticsPath, findingEnvelope(diagnosticsSchemaVersion, []any{findingGroup("advisory", "data_quality", "low_confidence", "normalize", 3)}, 4, true, 1))
+	rejections, err := loadRejections(rejectedPath)
+	if err != nil || len(rejections) != 1 || rejections[0].ReasonCode != "invalid" {
+		t.Fatalf("loadRejections() = %#v, %v", rejections, err)
+	}
+	warnings, err := loadWarnings(warningsPath)
+	if err != nil || len(warnings) != 1 || warnings[0].Category != "degradation" || warnings[0].OccurrenceCount != 2 {
+		t.Fatalf("loadWarnings() = %#v, %v", warnings, err)
+	}
+	diagnostics, occurrences, truncated, err := loadDiagnostics(diagnosticsPath)
+	if err != nil || len(diagnostics) != 1 || occurrences != 4 || !truncated || diagnostics[0].Category != "data_quality" {
+		t.Fatalf("loadDiagnostics() = %#v, %d, %t, %v", diagnostics, occurrences, truncated, err)
+	}
+
+	for name, path := range map[string]string{"rejections": rejectedPath, "warnings": warningsPath, "diagnostics": diagnosticsPath} {
+		t.Run("malformed "+name, func(t *testing.T) {
+			if err := os.WriteFile(path, []byte("{"), 0o644); err != nil {
+				t.Fatalf("WriteFile() error = %v", err)
+			}
+			var err error
+			if name == "rejections" {
+				_, err = loadRejections(path)
+			} else if name == "warnings" {
+				_, err = loadWarnings(path)
+			} else {
+				_, _, _, err = loadDiagnostics(path)
+			}
+			if err == nil {
+				t.Fatal("summary decoder error = nil")
+			}
+		})
+	}
+	oversized := filepath.Join(root, "oversized.json")
+	if err := os.WriteFile(oversized, []byte(strings.Repeat("x", maxSummaryBytes+1)), 0o644); err != nil {
+		t.Fatalf("WriteFile(oversized) error = %v", err)
+	}
+	if _, err := loadWarnings(oversized); err == nil || !strings.Contains(err.Error(), "exceeds") {
+		t.Fatalf("loadWarnings(oversized) error = %v", err)
+	}
+	if _, err := loadRejections(oversized); err == nil || !strings.Contains(err.Error(), "exceeds") {
+		t.Fatalf("loadRejections(oversized) error = %v", err)
+	}
+	if _, _, _, err := loadDiagnostics(oversized); err == nil || !strings.Contains(err.Error(), "exceeds") {
+		t.Fatalf("loadDiagnostics(oversized) error = %v", err)
+	}
+}
+
+func TestFakeRunnerCapturesRequestsAndHonorsContextAndError(t *testing.T) {
+	req := RunRequest{PipelineID: "pipeline", References: []ReferenceBinding{{Selector: "party", Path: "/references/party.json"}}}
+	want := RunResult{BundleRoot: "/bundle"}
+	fake := &FakeRunner{Result: want}
+	got, err := fake.Run(context.Background(), req)
+	if err != nil || !reflect.DeepEqual(got, want) || !reflect.DeepEqual(fake.Requests, []RunRequest{req}) {
+		t.Fatalf("Run() = %#v, %v; requests = %#v", got, err, fake.Requests)
+	}
+	req.References[0].Path = "/references/changed.json"
+	if fake.Requests[0].References[0].Path != "/references/party.json" {
+		t.Fatalf("fake retained aliased request references: %#v", fake.Requests[0].References)
+	}
+
+	wantErr := errors.New("configured failure")
+	fake.Err = wantErr
+	if _, err := fake.Run(context.Background(), req); !errors.Is(err, wantErr) {
+		t.Fatalf("Run(configured error) = %v", err)
+	}
+	canceled, cancel := context.WithCancel(context.Background())
+	cancel()
+	before := len(fake.Requests)
+	if _, err := fake.Run(canceled, req); !errors.Is(err, context.Canceled) || len(fake.Requests) != before {
+		t.Fatalf("Run(canceled) error = %v; requests = %d", err, len(fake.Requests))
+	}
+}
+
+func validRunRequest(t *testing.T) RunRequest {
+	t.Helper()
+	root := t.TempDir()
+	configPath := filepath.Join(root, "notarius.yml")
+	inputPath := filepath.Join(root, "input.json")
+	outputRoot := filepath.Join(root, "outputs")
+	workingDirectory := filepath.Join(root, "work")
+	diagnostics := filepath.Join(root, "diagnostics")
+	for _, directory := range []string{outputRoot, workingDirectory, diagnostics} {
+		if err := os.Mkdir(directory, 0o755); err != nil {
+			t.Fatalf("Mkdir(%q) error = %v", directory, err)
+		}
+	}
+	if err := os.WriteFile(configPath, []byte("pipelines: {}\n"), 0o644); err != nil {
+		t.Fatalf("WriteFile(config) error = %v", err)
+	}
+	if err := os.WriteFile(inputPath, []byte("{}\n"), 0o644); err != nil {
+		t.Fatalf("WriteFile(input) error = %v", err)
+	}
+	return RunRequest{
+		Binary: "notarius", ConfigPath: configPath, PipelineID: "dnd-session", InputPath: inputPath,
+		OutputRoot: outputRoot, WorkingDirectory: workingDirectory,
+		ReceiptPath: filepath.Join(diagnostics, "receipt.json"), LogPath: filepath.Join(diagnostics, "stderr.log"),
+		Timeout: time.Second,
+	}
+}
+
+func writeValidBundleAndReceipt(t *testing.T, req RunRequest, includeUnknown bool) {
+	t.Helper()
+	bundle := filepath.Join(req.OutputRoot, "notarius-run-1")
+	if err := os.MkdirAll(filepath.Join(bundle, "lanes"), 0o755); err != nil {
+		t.Fatalf("MkdirAll(bundle) error = %v", err)
+	}
+	for path, data := range map[string]string{
+		"manifest.json":         `{}`,
+		"lanes/npc.json":        `{}`,
+		"chunk-map.json":        `{}`,
+		"evidence-context.json": `{}`,
+	} {
+		if err := os.WriteFile(filepath.Join(bundle, filepath.FromSlash(path)), []byte(data), 0o644); err != nil {
+			t.Fatalf("WriteFile(%q) error = %v", path, err)
+		}
+	}
+	rejection := map[string]any{"stage": "validate", "lane_id": "spells", "reason_code": "invalid_spell", "message": strings.Repeat("external detail", 20)}
+	if includeUnknown {
+		rejection["future"] = true
+	}
+	writeJSONFile(t, filepath.Join(bundle, "rejected.json"), map[string]any{"rejected": []any{rejection}, "future": true})
+	writeJSONFile(t, filepath.Join(bundle, "warnings.json"), findingEnvelope(warningsSchemaVersion, []any{findingGroup("warning", "degradation", "normalized_name", "normalize", 2)}, 2, false, 0))
+	writeJSONFile(t, filepath.Join(bundle, "diagnostics.json"), findingEnvelope(diagnosticsSchemaVersion, []any{findingGroup("advisory", "data_quality", "low_confidence", "normalize", 3)}, 4, true, 1))
+	index := validIndexValue([]any{map[string]any{
+		"lane_id": "npc-registry", "file": "lanes/npc.json", "media_type": "application/json",
+		"module_key": "dnd/npc-registry", "schema_id": "notarius.dnd.npc_registry",
+		"schema_name": "NPCRegistry", "schema_version": "v1", "future": true,
+	}})
+	index["chunk_map"] = map[string]any{
+		"artifact_kind": "chunk_map", "file": "chunk-map.json", "media_type": "application/json",
+		"schema_id": "notarius.chunk_map", "schema_name": "ChunkMap", "schema_version": "v1", "future": true,
+	}
+	index["evidence_context"] = map[string]any{
+		"artifact_kind": "evidence_context", "file": "evidence-context.json", "media_type": "application/json",
+		"schema_id": "notarius.evidence_context", "schema_name": "EvidenceContext", "schema_version": "v1", "future": true,
+	}
+	index["future"] = true
+	writeJSONFile(t, filepath.Join(bundle, "index.json"), index)
+	receipt := map[string]any{
+		"schema_version": ReceiptSchemaVersion, "run_id": "notarius-run-1", "pipeline_id": req.PipelineID,
+		"output_directory": bundle, "index_file": "index.json", "normalized_output_count": 1,
+		"rejected_output_count": 1, "warning_group_count": 1, "warning_occurrence_count": 2,
+		"diagnostic_group_count": 1, "diagnostic_occurrence_count": 4,
+		"diagnostics_truncated": true, "validation_status": "rejected",
+		"validation_summaries": []any{map[string]any{
+			"stage": "normalize", "lane_id": "npc-registry", "status": "complete",
+			"producer_attempt_count": 1, "terminal_action": "accepted",
+		}},
+	}
+	if includeUnknown {
+		receipt["future"] = true
+	}
+	writeJSONFile(t, req.ReceiptPath, receipt)
+}
+
+func createBundleSkeleton(t *testing.T) string {
+	t.Helper()
+	bundle := filepath.Join(t.TempDir(), "bundle")
+	if err := os.MkdirAll(filepath.Join(bundle, "lanes"), 0o755); err != nil {
+		t.Fatalf("MkdirAll(bundle) error = %v", err)
+	}
+	for _, name := range []string{"manifest.json", "rejected.json", "warnings.json", "diagnostics.json", "lanes/npc.json", "chunk-map.json"} {
+		if err := os.WriteFile(filepath.Join(bundle, filepath.FromSlash(name)), []byte("{}"), 0o644); err != nil {
+			t.Fatalf("WriteFile(%q) error = %v", name, err)
+		}
+	}
+	return bundle
+}
+
+func validIndexValue(lanes []any) map[string]any {
+	return map[string]any{
+		"manifest_file": "manifest.json", "output_files": lanes,
+		"rejected_file": "rejected.json", "warnings_file": "warnings.json", "diagnostics_file": "diagnostics.json",
+	}
+}
+
+func findingGroup(disposition, category, reasonCode, origin string, occurrences int) map[string]any {
+	return map[string]any{
+		"disposition": disposition, "category": category, "reason_code": reasonCode,
+		"origin": map[string]any{"stage": origin, "lane_id": "npc-registry"}, "occurrence_count": occurrences,
+		"samples":              []any{map[string]any{"scope": "lane:npc-registry", "message": "external detail"}},
+		"omitted_sample_count": occurrences - 1,
+	}
+}
+
+func findingEnvelope(schema string, groups []any, occurrences int, truncated bool, unrepresented int) map[string]any {
+	value := map[string]any{
+		"schema_version": schema, "group_count": len(groups), "occurrence_count": occurrences,
+		"groups": groups,
+	}
+	if schema == diagnosticsSchemaVersion {
+		value["truncated"] = truncated
+		value["unrepresented_occurrence_count"] = unrepresented
+	}
+	return value
+}
+
+func writeJSONFile(t *testing.T, path string, value any) {
+	t.Helper()
+	data, err := json.Marshal(value)
+	if err != nil {
+		t.Fatalf("json.Marshal() error = %v", err)
+	}
+	if err := os.WriteFile(path, data, 0o644); err != nil {
+		t.Fatalf("WriteFile(%q) error = %v", path, err)
+	}
+}
+
+func writeShellScript(t *testing.T, body string) string {
+	t.Helper()
+	path := filepath.Join(t.TempDir(), "notarius-helper")
+	if err := os.WriteFile(path, []byte(body), 0o755); err != nil {
+		t.Fatalf("WriteFile(script) error = %v", err)
+	}
+	return path
+}
+
+func assertTextFile(t *testing.T, path, want string) {
+	t.Helper()
+	data, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatalf("ReadFile(%q) error = %v", path, err)
+	}
+	if string(data) != want {
+		t.Fatalf("ReadFile(%q) = %q, want %q", path, string(data), want)
+	}
+}
+
+func cloneMap(source map[string]any) map[string]any {
+	result := make(map[string]any, len(source))
+	for key, value := range source {
+		result[key] = value
+	}
+	return result
+}

internal/adapters/scriptorium/fake.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
 )
 
 // NoopRunner is a deterministic no-op scriptorium adapter.
@@ -142,7 +143,7 @@ func (f *FakeRunner) RenderArtifact(ctx context.Context, req RenderArtifactReque
 
 func materializeRunPlaceholders(req RunArtifactRequest) error {
 	if req.OutputPath != "" {
-		if err := subprocess.WriteFileAtomic(req.OutputPath, []byte("scriptorium noop/fake run artifact\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.OutputPath, []byte("scriptorium noop/fake run artifact\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write run output %q: %w", req.OutputPath, err)
 		}
 	}
@@ -154,17 +155,17 @@ func materializeRunPlaceholders(req RunArtifactRequest) error {
 			"prompt_id":   req.PromptID,
 			"output_path": req.OutputPath,
 		}
-		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644); err != nil {
+		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 	if req.StdoutLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("scriptorium noop/fake run stdout placeholder\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("scriptorium noop/fake run stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
 		}
 	}
 	if req.StderrLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("scriptorium noop/fake run stderr placeholder\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("scriptorium noop/fake run stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
 		}
 	}
@@ -173,7 +174,7 @@ func materializeRunPlaceholders(req RunArtifactRequest) error {
 
 func materializeRenderPlaceholders(req RenderArtifactRequest) error {
 	if req.OutputPath != "" {
-		if err := subprocess.WriteFileAtomic(req.OutputPath, []byte("{\"schema\":\"scriptorium.render.v1\",\"placeholder\":true}\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.OutputPath, []byte("{\"schema\":\"scriptorium.render.v1\",\"placeholder\":true}\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write render output %q: %w", req.OutputPath, err)
 		}
 	}
@@ -185,17 +186,17 @@ func materializeRenderPlaceholders(req RenderArtifactRequest) error {
 			"prompt_id":   req.PromptID,
 			"output_path": req.OutputPath,
 		}
-		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644); err != nil {
+		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 	if req.StdoutLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("scriptorium noop/fake render stdout placeholder\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("scriptorium noop/fake render stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
 		}
 	}
 	if req.StderrLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("scriptorium noop/fake render stderr placeholder\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("scriptorium noop/fake render stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
 		}
 	}

internal/adapters/scriptorium/subprocess.go

@@ -9,8 +9,12 @@ import (
 	"time"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
 )
 
+// MaxOutputFileBytes bounds one Scriptorium artifact result.
+const MaxOutputFileBytes int64 = 64 * 1024 * 1024
+
 // SubprocessRunner invokes Scriptorium through its public CLI.
 type SubprocessRunner struct{}
 
@@ -52,13 +56,17 @@ func (r *SubprocessRunner) RunArtifact(ctx context.Context, req RunArtifactReque
 		}
 	}
 
+	envOverrides, sensitiveNames := credentialEnvironment(req.APIKeyEnv)
 	runRes, runErr := subprocess.Run(ctx, subprocess.RunRequest{
-		Executable:    req.Binary,
-		Args:          args,
-		WorkingDir:    req.WorkingDir,
-		Timeout:       req.Timeout,
-		StdoutLogPath: req.StdoutLogPath,
-		StderrLogPath: req.StderrLogPath,
+		Executable:        req.Binary,
+		Args:              args,
+		WorkingDir:        req.WorkingDir,
+		Timeout:           req.Timeout,
+		EnvOverrides:      envOverrides,
+		SensitiveEnvNames: sensitiveNames,
+		DiagnosticOwner:   "scriptorium",
+		StdoutLogPath:     req.StdoutLogPath,
+		StderrLogPath:     req.StderrLogPath,
 	})
 
 	result := ArtifactResult{
@@ -134,13 +142,17 @@ func (r *SubprocessRunner) RenderArtifact(ctx context.Context, req RenderArtifac
 		}
 	}
 
+	envOverrides, sensitiveNames := credentialEnvironment(req.APIKeyEnv)
 	runRes, runErr := subprocess.Run(ctx, subprocess.RunRequest{
-		Executable:    req.Binary,
-		Args:          args,
-		WorkingDir:    req.WorkingDir,
-		Timeout:       req.Timeout,
-		StdoutLogPath: req.StdoutLogPath,
-		StderrLogPath: req.StderrLogPath,
+		Executable:        req.Binary,
+		Args:              args,
+		WorkingDir:        req.WorkingDir,
+		Timeout:           req.Timeout,
+		EnvOverrides:      envOverrides,
+		SensitiveEnvNames: sensitiveNames,
+		DiagnosticOwner:   "scriptorium",
+		StdoutLogPath:     req.StdoutLogPath,
+		StderrLogPath:     req.StderrLogPath,
 	})
 
 	result := ArtifactResult{
@@ -223,6 +235,15 @@ func validateCommonRunRequest(
 	return true, nil
 }
 
+func credentialEnvironment(apiKeyEnv string) (map[string]string, []string) {
+	name := strings.TrimSpace(apiKeyEnv)
+	if name == "" {
+		return nil, nil
+	}
+	value, _ := os.LookupEnv(name)
+	return map[string]string{name: value}, []string{name}
+}
+
 func buildRunArgs(req RunArtifactRequest) []string {
 	args := []string{"run", "--prompt", strings.TrimSpace(req.PromptID)}
 	if cfgPath := strings.TrimSpace(req.ConfigPath); cfgPath != "" {
@@ -321,18 +342,15 @@ func writeInvocationConfig(path string, payload invocationPayload) error {
 		"render_format":        payload.RenderFormat,
 		"render_prompt_logged": payload.RenderPromptStore,
 	}
-	return subprocess.WriteYAMLAtomic(path, data, 0o644)
+	return subprocess.WriteYAMLAtomic(path, data, fileops.WorkspaceFileMode)
 }
 
 func validateNonEmptyOutput(path string) error {
-	info, err := os.Stat(path)
+	data, err := fileops.ReadRegularFile(path, MaxOutputFileBytes)
 	if err != nil {
-		return fmt.Errorf("stat file: %w", err)
+		return fmt.Errorf("scriptorium artifact output exceeds or cannot be read within %d-byte limit: %w", MaxOutputFileBytes, err)
 	}
-	if info.IsDir() {
-		return fmt.Errorf("path is a directory")
-	}
-	if info.Size() <= 0 {
+	if len(data) == 0 {
 		return fmt.Errorf("file is empty")
 	}
 	return nil

internal/adapters/scriptorium/subprocess_test.go

@@ -31,7 +31,7 @@ func TestSubprocessRunnerRunSuccessBuildsDeterministicArgsAndCapturesLogs(t *tes
 		ConfigPath:          "/etc/scriptorium/config.yml",
 		PromptID:            "dnd.session_recap",
 		ProfileID:           "local-quality",
-		InputPaths:          map[string]string{"transcript": filepath.Join(dir, "processed.json"), "other": filepath.Join(dir, "other.md")},
+		InputPaths:          map[string]string{"transcript": filepath.Join(dir, "polished.json"), "other": filepath.Join(dir, "other.md")},
 		Vars:                map[string]string{"session_id": "2026-05-03", "campaign_name": "Icewind Dale"},
 		OutputPath:          filepath.Join(dir, "artifacts", "session_recap.md"),
 		StdoutLogPath:       filepath.Join(dir, "logs", "scriptorium.run.stdout.log"),
@@ -180,7 +180,7 @@ func TestSubprocessRunnerRenderSuccess(t *testing.T) {
 	req := RenderArtifactRequest{
 		Binary:              wrapper,
 		PromptID:            "dnd.session_recap",
-		InputPaths:          map[string]string{"transcript": filepath.Join(dir, "processed.json")},
+		InputPaths:          map[string]string{"transcript": filepath.Join(dir, "polished.json")},
 		OutputPath:          filepath.Join(dir, "artifacts", "session_recap.render.json"),
 		StdoutLogPath:       filepath.Join(dir, "logs", "scriptorium.render.stdout.log"),
 		StderrLogPath:       filepath.Join(dir, "logs", "scriptorium.render.stderr.log"),
@@ -285,7 +285,7 @@ type scriptoriumHelperRecord struct {
 func runReqForTest(t *testing.T, binary string) RunArtifactRequest {
 	t.Helper()
 	dir := t.TempDir()
-	transcriptPath := filepath.Join(dir, "processed.json")
+	transcriptPath := filepath.Join(dir, "polished.json")
 	writeScriptoriumFile(t, transcriptPath, `{"segments":[]}`)
 	return RunArtifactRequest{
 		Binary:              binary,

internal/adapters/seriatim/fake.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
 )
 
 // NoopRunner is a deterministic no-op seriatim adapter.
@@ -68,6 +69,26 @@ func (n *NoopRunner) Normalize(ctx context.Context, req NormalizeRequest) (Norma
 	}, nil
 }
 
+// Render returns the requested output path with placeholder metadata.
+func (n *NoopRunner) Render(ctx context.Context, req RenderRequest) (RenderResult, error) {
+	if err := ctx.Err(); err != nil {
+		return RenderResult{}, err
+	}
+	if err := materializeRenderPlaceholders(req); err != nil {
+		return RenderResult{}, err
+	}
+	return RenderResult{
+		OutputRenderedPath:  req.OutputRenderedPath,
+		StdoutLogPath:       req.StdoutLogPath,
+		StderrLogPath:       req.StderrLogPath,
+		GeneratedConfigPath: req.GeneratedConfigPath,
+		InvokedBinary:       "noop",
+		Format:              req.Format,
+		Title:               req.Title,
+		Metadata:            map[string]any{"placeholder": true},
+	}, nil
+}
+
 // FakeRunner captures merge requests and returns deterministic responses.
 type FakeRunner struct {
 	Requests          []MergeRequest
@@ -79,6 +100,9 @@ type FakeRunner struct {
 	TrimRequests      []TrimRequest
 	TrimErr           error
 	TrimResult        TrimResult
+	RenderRequests    []RenderRequest
+	RenderErr         error
+	RenderResult      RenderResult
 }
 
 // Run records request and returns configured response.
@@ -195,9 +219,49 @@ func (f *FakeRunner) Normalize(ctx context.Context, req NormalizeRequest) (Norma
 	return res, nil
 }
 
+// Render records request and returns configured response.
+func (f *FakeRunner) Render(ctx context.Context, req RenderRequest) (RenderResult, error) {
+	if err := ctx.Err(); err != nil {
+		return RenderResult{}, err
+	}
+	f.RenderRequests = append(f.RenderRequests, req)
+	if f.RenderErr != nil {
+		return RenderResult{}, f.RenderErr
+	}
+	if err := materializeRenderPlaceholders(req); err != nil {
+		return RenderResult{}, err
+	}
+	res := f.RenderResult
+	if res.OutputRenderedPath == "" {
+		res.OutputRenderedPath = req.OutputRenderedPath
+	}
+	if res.StdoutLogPath == "" {
+		res.StdoutLogPath = req.StdoutLogPath
+	}
+	if res.StderrLogPath == "" {
+		res.StderrLogPath = req.StderrLogPath
+	}
+	if res.GeneratedConfigPath == "" {
+		res.GeneratedConfigPath = req.GeneratedConfigPath
+	}
+	if res.InvokedBinary == "" {
+		res.InvokedBinary = "fake"
+	}
+	if res.Format == "" {
+		res.Format = req.Format
+	}
+	if res.Title == "" {
+		res.Title = req.Title
+	}
+	if res.Metadata == nil {
+		res.Metadata = map[string]any{"fake": true}
+	}
+	return res, nil
+}
+
 func materializePlaceholders(req MergeRequest) error {
 	if req.OutputMergedTranscriptPath != "" {
-		if err := subprocess.WriteFileAtomic(req.OutputMergedTranscriptPath, []byte(`{"schema":"seriatim.intermediate.v1","segments":[]}`), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.OutputMergedTranscriptPath, []byte(`{"schema":"seriatim.intermediate.v1","segments":[]}`), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write merged transcript %q: %w", req.OutputMergedTranscriptPath, err)
 		}
 	}
@@ -208,22 +272,22 @@ func materializePlaceholders(req MergeRequest) error {
 			"input_transcript_paths": req.InputTranscriptPaths,
 			"output_path":            req.OutputMergedTranscriptPath,
 		}
-		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644); err != nil {
+		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 	if req.StdoutLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake stdout placeholder\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
 		}
 	}
 	if req.StderrLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake stderr placeholder\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
 		}
 	}
 	if req.ReportPath != "" {
-		if err := subprocess.WriteFileAtomic(req.ReportPath, []byte(`{"schema":"seriatim.report.v1","placeholder":true}`), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.ReportPath, []byte(`{"schema":"seriatim.report.v1","placeholder":true}`), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write report %q: %w", req.ReportPath, err)
 		}
 	}
@@ -232,7 +296,7 @@ func materializePlaceholders(req MergeRequest) error {
 
 func materializeTrimPlaceholders(req TrimRequest) error {
 	if req.OutputTrimmedPath != "" {
-		if err := subprocess.WriteFileAtomic(req.OutputTrimmedPath, []byte(`{"schema":"seriatim.intermediate.v1","segments":[]}`), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.OutputTrimmedPath, []byte(`{"schema":"seriatim.intermediate.v1","segments":[]}`), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write trimmed transcript %q: %w", req.OutputTrimmedPath, err)
 		}
 	}
@@ -245,17 +309,17 @@ func materializeTrimPlaceholders(req TrimRequest) error {
 			"output_path":   req.OutputTrimmedPath,
 			"keep_selector": req.KeepSelector,
 		}
-		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644); err != nil {
+		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 	if req.StdoutLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake trim stdout placeholder\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake trim stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
 		}
 	}
 	if req.StderrLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake trim stderr placeholder\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake trim stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
 		}
 	}
@@ -264,7 +328,7 @@ func materializeTrimPlaceholders(req TrimRequest) error {
 
 func materializeNormalizePlaceholders(req NormalizeRequest) error {
 	if req.OutputNormalizedPath != "" {
-		if err := subprocess.WriteFileAtomic(req.OutputNormalizedPath, []byte(`{"schema":"seriatim.intermediate.v1","segments":[]}`), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.OutputNormalizedPath, []byte(`{"schema":"seriatim.intermediate.v1","segments":[]}`), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write normalized transcript %q: %w", req.OutputNormalizedPath, err)
 		}
 	}
@@ -280,24 +344,60 @@ func materializeNormalizePlaceholders(req NormalizeRequest) error {
 		if req.ReportPath != "" {
 			payload["report_path"] = req.ReportPath
 		}
-		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644); err != nil {
+		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
 		}
 	}
 	if req.StdoutLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake normalize stdout placeholder\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake normalize stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
 		}
 	}
 	if req.StderrLogPath != "" {
-		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake normalize stderr placeholder\n"), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake normalize stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
 		}
 	}
 	if req.ReportPath != "" {
-		if err := subprocess.WriteFileAtomic(req.ReportPath, []byte(`{"schema":"seriatim.report.v1","placeholder":true}`), 0o644); err != nil {
+		if err := subprocess.WriteFileAtomic(req.ReportPath, []byte(`{"schema":"seriatim.report.v1","placeholder":true}`), fileops.WorkspaceFileMode); err != nil {
 			return fmt.Errorf("write report %q: %w", req.ReportPath, err)
 		}
 	}
 	return nil
 }
+
+func materializeRenderPlaceholders(req RenderRequest) error {
+	if req.OutputRenderedPath != "" {
+		if err := subprocess.WriteFileAtomic(req.OutputRenderedPath, []byte("# Transcript\n\nRendered markdown placeholder.\n"), fileops.WorkspaceFileMode); err != nil {
+			return fmt.Errorf("write rendered transcript %q: %w", req.OutputRenderedPath, err)
+		}
+	}
+	if req.GeneratedConfigPath != "" {
+		payload := map[string]any{
+			"schema":              "seriatim.generated.v1",
+			"placeholder":         true,
+			"command":             "render",
+			"input_path":          req.InputTranscriptPath,
+			"output_path":         req.OutputRenderedPath,
+			"format":              req.Format,
+			"title":               req.Title,
+			"include_timestamps":  req.IncludeTimestamps,
+			"include_segment_ids": req.IncludeSegmentIDs,
+			"include_metadata":    req.IncludeMetadata,
+		}
+		if err := subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, fileops.WorkspaceFileMode); err != nil {
+			return fmt.Errorf("write generated config %q: %w", req.GeneratedConfigPath, err)
+		}
+	}
+	if req.StdoutLogPath != "" {
+		if err := subprocess.WriteFileAtomic(req.StdoutLogPath, []byte("seriatim noop/fake render stdout placeholder\n"), fileops.WorkspaceFileMode); err != nil {
+			return fmt.Errorf("write stdout log %q: %w", req.StdoutLogPath, err)
+		}
+	}
+	if req.StderrLogPath != "" {
+		if err := subprocess.WriteFileAtomic(req.StderrLogPath, []byte("seriatim noop/fake render stderr placeholder\n"), fileops.WorkspaceFileMode); err != nil {
+			return fmt.Errorf("write stderr log %q: %w", req.StderrLogPath, err)
+		}
+	}
+	return nil
+}

internal/adapters/seriatim/fake_test.go

@@ -14,7 +14,7 @@ func TestFakeRunnerCapturesRequestAndReturnsPath(t *testing.T) {
 	dir := t.TempDir()
 	req := MergeRequest{
 		GeneratedConfigPath:        filepath.Join(dir, "config", "seriatim.yml"),
-		OutputMergedTranscriptPath: filepath.Join(dir, "transcripts", "merged.json"),
+		OutputMergedTranscriptPath: filepath.Join(dir, "transcripts", "base.json"),
 		StdoutLogPath:              filepath.Join(dir, "logs", "seriatim.stdout.log"),
 		StderrLogPath:              filepath.Join(dir, "logs", "seriatim.stderr.log"),
 	}
@@ -57,8 +57,8 @@ func TestFakeRunnerTrimCapturesRequestAndReturnsPath(t *testing.T) {
 	dir := t.TempDir()
 	req := TrimRequest{
 		GeneratedConfigPath: filepath.Join(dir, "config", "seriatim.trim.yml"),
-		InputTranscriptPath: filepath.Join(dir, "transcripts", "processed.json"),
-		OutputTrimmedPath:   filepath.Join(dir, "transcripts", "trimmed.json"),
+		InputTranscriptPath: filepath.Join(dir, "transcripts", "polished.json"),
+		OutputTrimmedPath:   filepath.Join(dir, "transcripts", "final.trimmed.json"),
 		KeepSelector:        "1-10",
 		StdoutLogPath:       filepath.Join(dir, "logs", "seriatim.trim.stdout.log"),
 		StderrLogPath:       filepath.Join(dir, "logs", "seriatim.trim.stderr.log"),
@@ -105,8 +105,8 @@ func TestFakeRunnerNormalizeCapturesRequestAndReturnsPath(t *testing.T) {
 	dir := t.TempDir()
 	req := NormalizeRequest{
 		GeneratedConfigPath:  filepath.Join(dir, "config", "seriatim.normalize.yml"),
-		InputTranscriptPath:  filepath.Join(dir, "transcripts", "processed.json"),
-		OutputNormalizedPath: filepath.Join(dir, "transcripts", "normalized.json"),
+		InputTranscriptPath:  filepath.Join(dir, "transcripts", "polished.json"),
+		OutputNormalizedPath: filepath.Join(dir, "transcripts", "final.json"),
 		OutputSchema:         "seriatim-intermediate",
 		ReportPath:           filepath.Join(dir, "artifacts", "seriatim.normalize.report.json"),
 		StdoutLogPath:        filepath.Join(dir, "logs", "seriatim.normalize.stdout.log"),
@@ -148,3 +148,58 @@ func TestFakeRunnerNormalizeError(t *testing.T) {
 		t.Fatal("expected error, got nil")
 	}
 }
+
+func TestFakeRunnerRenderCapturesRequestAndReturnsPath(t *testing.T) {
+	fake := &FakeRunner{}
+	dir := t.TempDir()
+	req := RenderRequest{
+		GeneratedConfigPath: filepath.Join(dir, "config", "seriatim.render.yml"),
+		InputTranscriptPath: filepath.Join(dir, "transcripts", "final.trimmed.json"),
+		OutputRenderedPath:  filepath.Join(dir, "transcripts", "final.trimmed.md"),
+		Format:              "markdown",
+		Title:               "Session render",
+		IncludeTimestamps:   true,
+		IncludeSegmentIDs:   false,
+		IncludeMetadata:     true,
+		StdoutLogPath:       filepath.Join(dir, "logs", "seriatim.render.stdout.log"),
+		StderrLogPath:       filepath.Join(dir, "logs", "seriatim.render.stderr.log"),
+	}
+
+	res, err := fake.Render(context.Background(), req)
+	if err != nil {
+		t.Fatalf("Render() error = %v", err)
+	}
+	if len(fake.RenderRequests) != 1 || fake.RenderRequests[0].GeneratedConfigPath == "" {
+		t.Fatalf("render requests = %#v, want captured request", fake.RenderRequests)
+	}
+	if res.OutputRenderedPath != req.OutputRenderedPath {
+		t.Fatalf("rendered path = %q, want %q", res.OutputRenderedPath, req.OutputRenderedPath)
+	}
+	if res.Format != req.Format {
+		t.Fatalf("format = %q, want %q", res.Format, req.Format)
+	}
+	if res.Title != req.Title {
+		t.Fatalf("title = %q, want %q", res.Title, req.Title)
+	}
+
+	cfgData, err := os.ReadFile(req.GeneratedConfigPath)
+	if err != nil {
+		t.Fatalf("read generated config: %v", err)
+	}
+	if !strings.Contains(string(cfgData), "command: render") {
+		t.Fatalf("generated config = %q, want render command marker", string(cfgData))
+	}
+	for _, path := range []string{req.StdoutLogPath, req.StderrLogPath, req.OutputRenderedPath} {
+		if _, err := os.Stat(path); err != nil {
+			t.Fatalf("expected file %q to exist: %v", path, err)
+		}
+	}
+}
+
+func TestFakeRunnerRenderError(t *testing.T) {
+	fake := &FakeRunner{RenderErr: errors.New("boom")}
+	_, err := fake.Render(context.Background(), RenderRequest{})
+	if err == nil {
+		t.Fatal("expected error, got nil")
+	}
+}

internal/adapters/seriatim/runner.go

@@ -1,4 +1,4 @@
-// Package seriatim declares the adapter contract for transcript merge/normalize/trim execution.
+// Package seriatim declares the adapter contract for transcript merge/normalize/trim/render execution.
 package seriatim
 
 import (
@@ -6,11 +6,12 @@ import (
 	"time"
 )
 
-// Runner is the adapter boundary for seriatim merge/normalize/trim invocations.
+// Runner is the adapter boundary for seriatim merge/normalize/trim/render invocations.
 type Runner interface {
 	Run(ctx context.Context, req MergeRequest) (MergeResult, error)
 	Normalize(ctx context.Context, req NormalizeRequest) (NormalizeResult, error)
 	Trim(ctx context.Context, req TrimRequest) (TrimResult, error)
+	Render(ctx context.Context, req RenderRequest) (RenderResult, error)
 }
 
 // MergeRequest describes a seriatim merge invocation.
@@ -90,3 +91,33 @@ type TrimResult struct {
 	KeepSelector        string
 	Metadata            map[string]any
 }
+
+// RenderRequest describes a seriatim render invocation.
+type RenderRequest struct {
+	Binary              string
+	InputTranscriptPath string
+	OutputRenderedPath  string
+	Format              string
+	Title               string
+	IncludeTimestamps   bool
+	IncludeSegmentIDs   bool
+	IncludeMetadata     bool
+	StdoutLogPath       string
+	StderrLogPath       string
+	GeneratedConfigPath string
+	Timeout             time.Duration
+}
+
+// RenderResult describes a render output.
+type RenderResult struct {
+	OutputRenderedPath  string
+	StdoutLogPath       string
+	StderrLogPath       string
+	GeneratedConfigPath string
+	ExitCode            int
+	Duration            time.Duration
+	InvokedBinary       string
+	Format              string
+	Title               string
+	Metadata            map[string]any
+}

internal/adapters/seriatim/subprocess.go

@@ -4,14 +4,18 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
-	"os"
 	"strconv"
 	"strings"
 	"time"
+	"unicode/utf8"
 
 	"gitea.maximumdirect.net/eric/narratio/internal/adapters/subprocess"
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
 )
 
+// MaxOutputFileBytes bounds each Seriatim JSON or rendered-text result.
+const MaxOutputFileBytes int64 = 64 * 1024 * 1024
+
 // EnvConfig defines optional Seriatim environment tuning values.
 type EnvConfig struct {
 	OverlapWordRunGap           *float64
@@ -128,12 +132,13 @@ func (r *SubprocessRunner) Run(ctx context.Context, req MergeRequest) (MergeResu
 	}
 
 	runRes, err := subprocess.Run(ctx, subprocess.RunRequest{
-		Executable:    r.binary,
-		Args:          args,
-		Timeout:       r.timeout,
-		EnvOverrides:  env,
-		StdoutLogPath: req.StdoutLogPath,
-		StderrLogPath: req.StderrLogPath,
+		Executable:      r.binary,
+		Args:            args,
+		Timeout:         r.timeout,
+		EnvOverrides:    env,
+		DiagnosticOwner: "seriatim",
+		StdoutLogPath:   req.StdoutLogPath,
+		StderrLogPath:   req.StderrLogPath,
 	})
 	if err != nil {
 		return MergeResult{
@@ -231,11 +236,12 @@ func (r *SubprocessRunner) Trim(ctx context.Context, req TrimRequest) (TrimResul
 	}
 
 	runRes, err := subprocess.Run(ctx, subprocess.RunRequest{
-		Executable:    binary,
-		Args:          args,
-		Timeout:       timeout,
-		StdoutLogPath: req.StdoutLogPath,
-		StderrLogPath: req.StderrLogPath,
+		Executable:      binary,
+		Args:            args,
+		Timeout:         timeout,
+		DiagnosticOwner: "seriatim",
+		StdoutLogPath:   req.StdoutLogPath,
+		StderrLogPath:   req.StderrLogPath,
 	})
 	if err != nil {
 		return TrimResult{
@@ -319,11 +325,12 @@ func (r *SubprocessRunner) Normalize(ctx context.Context, req NormalizeRequest)
 	}
 
 	runRes, err := subprocess.Run(ctx, subprocess.RunRequest{
-		Executable:    binary,
-		Args:          args,
-		Timeout:       timeout,
-		StdoutLogPath: req.StdoutLogPath,
-		StderrLogPath: req.StderrLogPath,
+		Executable:      binary,
+		Args:            args,
+		Timeout:         timeout,
+		DiagnosticOwner: "seriatim",
+		StdoutLogPath:   req.StdoutLogPath,
+		StderrLogPath:   req.StderrLogPath,
 	})
 	if err != nil {
 		return NormalizeResult{
@@ -384,6 +391,97 @@ func (r *SubprocessRunner) Normalize(ctx context.Context, req NormalizeRequest)
 	}, nil
 }
 
+// Render executes Seriatim render with deterministic flags and validates non-empty text output.
+func (r *SubprocessRunner) Render(ctx context.Context, req RenderRequest) (RenderResult, error) {
+	if r == nil {
+		return RenderResult{}, fmt.Errorf("seriatim subprocess runner is nil")
+	}
+	if strings.TrimSpace(req.InputTranscriptPath) == "" {
+		return RenderResult{}, fmt.Errorf("seriatim render input path is required")
+	}
+	if strings.TrimSpace(req.OutputRenderedPath) == "" {
+		return RenderResult{}, fmt.Errorf("seriatim render output path is required")
+	}
+	format := strings.TrimSpace(req.Format)
+	if format == "" {
+		format = "markdown"
+	}
+	if format != "markdown" {
+		return RenderResult{}, fmt.Errorf("seriatim render format %q is unsupported", req.Format)
+	}
+
+	binary := r.binary
+	if strings.TrimSpace(req.Binary) != "" {
+		binary = strings.TrimSpace(req.Binary)
+	}
+
+	timeout := r.timeout
+	if req.Timeout < 0 {
+		return RenderResult{}, fmt.Errorf("seriatim render timeout must be >= 0")
+	}
+	if req.Timeout > 0 {
+		timeout = req.Timeout
+	}
+
+	args := buildRenderArgs(req, format)
+	if req.GeneratedConfigPath != "" {
+		if err := writeRenderInvocationConfig(req, args, binary, timeout, format); err != nil {
+			return RenderResult{}, fmt.Errorf("write seriatim render invocation config %q: %w", req.GeneratedConfigPath, err)
+		}
+	}
+
+	runRes, err := subprocess.Run(ctx, subprocess.RunRequest{
+		Executable:      binary,
+		Args:            args,
+		Timeout:         timeout,
+		DiagnosticOwner: "seriatim",
+		StdoutLogPath:   req.StdoutLogPath,
+		StderrLogPath:   req.StderrLogPath,
+	})
+	if err != nil {
+		return RenderResult{
+			OutputRenderedPath:  req.OutputRenderedPath,
+			StdoutLogPath:       req.StdoutLogPath,
+			StderrLogPath:       req.StderrLogPath,
+			GeneratedConfigPath: req.GeneratedConfigPath,
+			ExitCode:            runRes.ExitCode,
+			Duration:            runRes.Duration,
+			InvokedBinary:       binary,
+			Format:              format,
+			Title:               req.Title,
+		}, fmt.Errorf("run seriatim render (binary=%q): %w", binary, err)
+	}
+
+	if err := validateNonEmptyTextFile(req.OutputRenderedPath); err != nil {
+		return RenderResult{
+			OutputRenderedPath:  req.OutputRenderedPath,
+			StdoutLogPath:       req.StdoutLogPath,
+			StderrLogPath:       req.StderrLogPath,
+			GeneratedConfigPath: req.GeneratedConfigPath,
+			ExitCode:            runRes.ExitCode,
+			Duration:            runRes.Duration,
+			InvokedBinary:       binary,
+			Format:              format,
+			Title:               req.Title,
+		}, fmt.Errorf("validate seriatim rendered output %q: %w", req.OutputRenderedPath, err)
+	}
+
+	return RenderResult{
+		OutputRenderedPath:  req.OutputRenderedPath,
+		StdoutLogPath:       req.StdoutLogPath,
+		StderrLogPath:       req.StderrLogPath,
+		GeneratedConfigPath: req.GeneratedConfigPath,
+		ExitCode:            runRes.ExitCode,
+		Duration:            runRes.Duration,
+		InvokedBinary:       binary,
+		Format:              format,
+		Title:               req.Title,
+		Metadata: map[string]any{
+			"adapter": "seriatim_subprocess",
+		},
+	}, nil
+}
+
 func (r *SubprocessRunner) buildMergeArgs(req MergeRequest) []string {
 	args := []string{"merge"}
 
@@ -455,7 +553,7 @@ func (r *SubprocessRunner) writeMergeInvocationConfig(req MergeRequest, args []s
 		payload["coalesce_gap"] = *r.coalesceGap
 	}
 
-	return subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644)
+	return subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, fileops.WorkspaceFileMode)
 }
 
 func buildTrimArgs(req TrimRequest) []string {
@@ -480,6 +578,22 @@ func buildNormalizeArgs(req NormalizeRequest, outputSchema string) []string {
 	return args
 }
 
+func buildRenderArgs(req RenderRequest, format string) []string {
+	args := []string{
+		"render",
+		"--input-file", req.InputTranscriptPath,
+		"--output-file", req.OutputRenderedPath,
+		"--format", format,
+		"--include-timestamps=" + strconv.FormatBool(req.IncludeTimestamps),
+		"--include-segment-ids=" + strconv.FormatBool(req.IncludeSegmentIDs),
+		"--include-metadata=" + strconv.FormatBool(req.IncludeMetadata),
+	}
+	if strings.TrimSpace(req.Title) != "" {
+		args = append(args, "--title", req.Title)
+	}
+	return args
+}
+
 func writeTrimInvocationConfig(req TrimRequest, args []string, binary string, timeout time.Duration) error {
 	payload := map[string]any{
 		"schema":        "seriatim.generated.v1",
@@ -491,7 +605,7 @@ func writeTrimInvocationConfig(req TrimRequest, args []string, binary string, ti
 		"output_path":   req.OutputTrimmedPath,
 		"keep_selector": req.KeepSelector,
 	}
-	return subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644)
+	return subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, fileops.WorkspaceFileMode)
 }
 
 func writeNormalizeInvocationConfig(req NormalizeRequest, args []string, binary string, timeout time.Duration, outputSchema string) error {
@@ -506,13 +620,31 @@ func writeNormalizeInvocationConfig(req NormalizeRequest, args []string, binary
 		"output_schema": outputSchema,
 		"report_path":   req.ReportPath,
 	}
-	return subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, 0o644)
+	return subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, fileops.WorkspaceFileMode)
+}
+
+func writeRenderInvocationConfig(req RenderRequest, args []string, binary string, timeout time.Duration, format string) error {
+	payload := map[string]any{
+		"schema":              "seriatim.generated.v1",
+		"command":             "render",
+		"binary":              binary,
+		"args":                args,
+		"timeout":             timeout.String(),
+		"input_path":          req.InputTranscriptPath,
+		"output_path":         req.OutputRenderedPath,
+		"format":              format,
+		"title":               req.Title,
+		"include_timestamps":  req.IncludeTimestamps,
+		"include_segment_ids": req.IncludeSegmentIDs,
+		"include_metadata":    req.IncludeMetadata,
+	}
+	return subprocess.WriteYAMLAtomic(req.GeneratedConfigPath, payload, fileops.WorkspaceFileMode)
 }
 
 func validateJSONFile(path string) error {
-	data, err := os.ReadFile(path)
+	data, err := readSeriatimResult(path, "JSON output")
 	if err != nil {
-		return fmt.Errorf("read file: %w", err)
+		return err
 	}
 	var v any
 	if err := json.Unmarshal(data, &v); err != nil {
@@ -522,9 +654,9 @@ func validateJSONFile(path string) error {
 }
 
 func validateJSONFileWithSegments(path string) error {
-	data, err := os.ReadFile(path)
+	data, err := readSeriatimResult(path, "transcript JSON output")
 	if err != nil {
-		return fmt.Errorf("read file: %w", err)
+		return err
 	}
 
 	var payload map[string]any
@@ -541,3 +673,28 @@ func validateJSONFileWithSegments(path string) error {
 	}
 	return nil
 }
+
+func validateNonEmptyTextFile(path string) error {
+	data, err := readSeriatimResult(path, "rendered text output")
+	if err != nil {
+		return err
+	}
+	if len(data) == 0 {
+		return fmt.Errorf("file is empty")
+	}
+	if !utf8.Valid(data) {
+		return fmt.Errorf("file is not valid utf-8 text")
+	}
+	if strings.TrimSpace(string(data)) == "" {
+		return fmt.Errorf("file has no non-whitespace content")
+	}
+	return nil
+}
+
+func readSeriatimResult(path, category string) ([]byte, error) {
+	data, err := fileops.ReadRegularFile(path, MaxOutputFileBytes)
+	if err != nil {
+		return nil, fmt.Errorf("seriatim %s exceeds or cannot be read within %d-byte limit: %w", category, MaxOutputFileBytes, err)
+	}
+	return data, nil
+}

internal/adapters/seriatim/subprocess_test.go

@@ -50,7 +50,7 @@ func TestSubprocessRunnerSuccessWithReportArgsAndEnv(t *testing.T) {
 	req := MergeRequest{
 		GeneratedConfigPath:        filepath.Join(dir, "seriatim.generated.yml"),
 		InputTranscriptPaths:       []string{filepath.Join(dir, "a.json"), filepath.Join(dir, "b.json")},
-		OutputMergedTranscriptPath: filepath.Join(dir, "merged.json"),
+		OutputMergedTranscriptPath: filepath.Join(dir, "base.json"),
 		ReportPath:                 filepath.Join(dir, "seriatim.report.json"),
 		SpeakersPath:               filepath.Join(dir, "speakers.yml"),
 		AutocorrectPath:            filepath.Join(dir, "autocorrect.yml"),
@@ -569,6 +569,156 @@ func TestSubprocessRunnerNormalizeInvalidReportJSONFails(t *testing.T) {
 	}
 }
 
+func TestSubprocessRunnerRenderSuccessInvocationAndProvenance(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("helper wrapper script uses /bin/sh")
+	}
+
+	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
+	t.Setenv("SERIATIM_HELPER_MODE", "render_success")
+	recordPath := filepath.Join(t.TempDir(), "record.json")
+	t.Setenv("SERIATIM_HELPER_RECORD_PATH", recordPath)
+
+	wrapper := writeHelperWrapper(t)
+	runner := mustRunner(t, wrapper, false)
+	req := renderReqForTest(t)
+
+	res, err := runner.Render(context.Background(), req)
+	if err != nil {
+		t.Fatalf("Render() error = %v", err)
+	}
+	if res.OutputRenderedPath != req.OutputRenderedPath {
+		t.Fatalf("OutputRenderedPath = %q, want %q", res.OutputRenderedPath, req.OutputRenderedPath)
+	}
+	if res.Format != req.Format {
+		t.Fatalf("Format = %q, want %q", res.Format, req.Format)
+	}
+	if res.Title != req.Title {
+		t.Fatalf("Title = %q, want %q", res.Title, req.Title)
+	}
+	if res.InvokedBinary != wrapper {
+		t.Fatalf("InvokedBinary = %q, want %q", res.InvokedBinary, wrapper)
+	}
+	if res.ExitCode != 0 {
+		t.Fatalf("ExitCode = %d, want 0", res.ExitCode)
+	}
+	if res.Duration <= 0 {
+		t.Fatalf("Duration = %s, want >0", res.Duration)
+	}
+	if res.Metadata == nil || res.Metadata["adapter"] != "seriatim_subprocess" {
+		t.Fatalf("Metadata = %#v, want adapter marker", res.Metadata)
+	}
+
+	if _, err := os.Stat(req.OutputRenderedPath); err != nil {
+		t.Fatalf("rendered output missing: %v", err)
+	}
+	if _, err := os.Stat(req.StdoutLogPath); err != nil {
+		t.Fatalf("stdout log missing: %v", err)
+	}
+	if _, err := os.Stat(req.StderrLogPath); err != nil {
+		t.Fatalf("stderr log missing: %v", err)
+	}
+	if _, err := os.Stat(req.GeneratedConfigPath); err != nil {
+		t.Fatalf("generated config missing: %v", err)
+	}
+
+	rec := readHelperRecord(t, recordPath)
+	wantArgs := []string{
+		"render",
+		"--input-file", req.InputTranscriptPath,
+		"--output-file", req.OutputRenderedPath,
+		"--format", req.Format,
+		"--include-timestamps=true",
+		"--include-segment-ids=true",
+		"--include-metadata=false",
+		"--title", req.Title,
+	}
+	if strings.Join(rec.Args, "\n") != strings.Join(wantArgs, "\n") {
+		t.Fatalf("args = %#v, want %#v", rec.Args, wantArgs)
+	}
+}
+
+func TestSubprocessRunnerRenderWithoutTitleOmitsTitleArg(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("helper wrapper script uses /bin/sh")
+	}
+	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
+	t.Setenv("SERIATIM_HELPER_MODE", "render_success")
+	recordPath := filepath.Join(t.TempDir(), "record.json")
+	t.Setenv("SERIATIM_HELPER_RECORD_PATH", recordPath)
+
+	runner := mustRunner(t, writeHelperWrapper(t), false)
+	req := renderReqForTest(t)
+	req.Title = ""
+	if _, err := runner.Render(context.Background(), req); err != nil {
+		t.Fatalf("Render() error = %v", err)
+	}
+
+	rec := readHelperRecord(t, recordPath)
+	for i := 0; i < len(rec.Args); i++ {
+		if rec.Args[i] == "--title" {
+			t.Fatalf("args = %#v, did not expect --title", rec.Args)
+		}
+	}
+}
+
+func TestSubprocessRunnerRenderSubprocessFailure(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("helper wrapper script uses /bin/sh")
+	}
+	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
+	t.Setenv("SERIATIM_HELPER_MODE", "fail")
+	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
+
+	runner := mustRunner(t, writeHelperWrapper(t), false)
+	req := renderReqForTest(t)
+	_, err := runner.Render(context.Background(), req)
+	if err == nil {
+		t.Fatal("Render() error = nil, want non-nil")
+	}
+	if !strings.Contains(err.Error(), "run seriatim render") {
+		t.Fatalf("error = %q, want subprocess context", err.Error())
+	}
+}
+
+func TestSubprocessRunnerRenderMissingOutputFails(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("helper wrapper script uses /bin/sh")
+	}
+	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
+	t.Setenv("SERIATIM_HELPER_MODE", "missing_output")
+	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
+
+	runner := mustRunner(t, writeHelperWrapper(t), false)
+	req := renderReqForTest(t)
+	_, err := runner.Render(context.Background(), req)
+	if err == nil {
+		t.Fatal("Render() error = nil, want non-nil")
+	}
+	if !strings.Contains(err.Error(), "validate seriatim rendered output") {
+		t.Fatalf("error = %q, want output validation context", err.Error())
+	}
+}
+
+func TestSubprocessRunnerRenderEmptyOutputFails(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("helper wrapper script uses /bin/sh")
+	}
+	t.Setenv("GO_WANT_SERIATIM_HELPER", "1")
+	t.Setenv("SERIATIM_HELPER_MODE", "render_empty_output")
+	t.Setenv("SERIATIM_HELPER_RECORD_PATH", filepath.Join(t.TempDir(), "record.json"))
+
+	runner := mustRunner(t, writeHelperWrapper(t), false)
+	req := renderReqForTest(t)
+	_, err := runner.Render(context.Background(), req)
+	if err == nil {
+		t.Fatal("Render() error = nil, want non-nil")
+	}
+	if !strings.Contains(err.Error(), "file is empty") {
+		t.Fatalf("error = %q, want empty-file validation", err.Error())
+	}
+}
+
 func TestSubprocessRunnerConstructorValidation(t *testing.T) {
 	_, err := NewSubprocessRunnerFromConfigValues("", "10m", "seriatim-intermediate", nil, true, EnvConfig{})
 	if err == nil {
@@ -702,6 +852,14 @@ func TestSeriatimSubprocessHelper(t *testing.T) {
 	case "normalize_report_missing":
 		writeSeriatimHelperFile(outputPath, `{"schema":"seriatim.intermediate.v1","segments":[]}`)
 		os.Exit(0)
+	case "render_success":
+		writeSeriatimHelperFile(outputPath, "# Rendered transcript\n\nHello.\n")
+		_, _ = os.Stdout.WriteString("seriatim helper render stdout\n")
+		_, _ = os.Stderr.WriteString("seriatim helper render stderr\n")
+		os.Exit(0)
+	case "render_empty_output":
+		writeSeriatimHelperFile(outputPath, "")
+		os.Exit(0)
 	default:
 		_, _ = os.Stderr.WriteString(fmt.Sprintf("unknown helper mode %q\n", mode))
 		os.Exit(2)
@@ -732,7 +890,7 @@ func mergeReqForTest(t *testing.T, withReport bool) MergeRequest {
 	req := MergeRequest{
 		GeneratedConfigPath:        filepath.Join(dir, "seriatim.generated.yml"),
 		InputTranscriptPaths:       []string{in1, in2},
-		OutputMergedTranscriptPath: filepath.Join(dir, "merged.json"),
+		OutputMergedTranscriptPath: filepath.Join(dir, "base.json"),
 		StdoutLogPath:              filepath.Join(dir, "seriatim.stdout.log"),
 		StderrLogPath:              filepath.Join(dir, "seriatim.stderr.log"),
 	}
@@ -745,11 +903,11 @@ func mergeReqForTest(t *testing.T, withReport bool) MergeRequest {
 func trimReqForTest(t *testing.T) TrimRequest {
 	t.Helper()
 	dir := t.TempDir()
-	input := filepath.Join(dir, "processed.json")
+	input := filepath.Join(dir, "polished.json")
 	writeSeriatimFile(t, input, `{"schema":"seriatim.intermediate.v1","segments":[]}`)
 	return TrimRequest{
 		InputTranscriptPath: input,
-		OutputTrimmedPath:   filepath.Join(dir, "trimmed.json"),
+		OutputTrimmedPath:   filepath.Join(dir, "final.trimmed.json"),
 		KeepSelector:        "5-12",
 		GeneratedConfigPath: filepath.Join(dir, "seriatim.trim.generated.yml"),
 		StdoutLogPath:       filepath.Join(dir, "seriatim.trim.stdout.log"),
@@ -760,12 +918,12 @@ func trimReqForTest(t *testing.T) TrimRequest {
 func normalizeReqForTest(t *testing.T, withReport bool) NormalizeRequest {
 	t.Helper()
 	dir := t.TempDir()
-	input := filepath.Join(dir, "processed.json")
+	input := filepath.Join(dir, "polished.json")
 	writeSeriatimFile(t, input, `{"schema":"audita.processed.v1","segments":[]}`)
 
 	req := NormalizeRequest{
 		InputTranscriptPath:  input,
-		OutputNormalizedPath: filepath.Join(dir, "normalized.json"),
+		OutputNormalizedPath: filepath.Join(dir, "final.json"),
 		OutputSchema:         "seriatim-intermediate",
 		GeneratedConfigPath:  filepath.Join(dir, "seriatim.normalize.generated.yml"),
 		StdoutLogPath:        filepath.Join(dir, "seriatim.normalize.stdout.log"),
@@ -777,6 +935,25 @@ func normalizeReqForTest(t *testing.T, withReport bool) NormalizeRequest {
 	return req
 }
 
+func renderReqForTest(t *testing.T) RenderRequest {
+	t.Helper()
+	dir := t.TempDir()
+	input := filepath.Join(dir, "final.trimmed.json")
+	writeSeriatimFile(t, input, `{"schema":"seriatim.intermediate.v1","segments":[]}`)
+	return RenderRequest{
+		InputTranscriptPath: input,
+		OutputRenderedPath:  filepath.Join(dir, "final.trimmed.md"),
+		Format:              "markdown",
+		Title:               "Session 42",
+		IncludeTimestamps:   true,
+		IncludeSegmentIDs:   true,
+		IncludeMetadata:     false,
+		GeneratedConfigPath: filepath.Join(dir, "seriatim.render.generated.yml"),
+		StdoutLogPath:       filepath.Join(dir, "seriatim.render.stdout.log"),
+		StderrLogPath:       filepath.Join(dir, "seriatim.render.stderr.log"),
+	}
+}
+
 func mustRunner(t *testing.T, binary string, report bool) *SubprocessRunner {
 	t.Helper()
 	coalesce := 3.0

internal/adapters/storage/archive.go

@@ -1,31 +0,0 @@
-// Package storage declares archive/storage backend adapter boundaries.
-package storage
-
-import "context"
-
-// TODO: implement remote storage/archive backends (S3/SFTP/etc.).
-
-// Backend is the adapter boundary for archive/storage operations.
-type Backend interface {
-	Archive(ctx context.Context, req ArchiveRequest) (ArchiveResult, error)
-}
-
-// ArchiveItem describes one item to archive.
-type ArchiveItem struct {
-	Kind      string
-	LocalPath string
-	RemoteKey string
-}
-
-// ArchiveRequest describes one archive operation.
-type ArchiveRequest struct {
-	SessionID    string
-	ManifestPath string
-	Items        []ArchiveItem
-}
-
-// ArchiveResult describes archive operation output.
-type ArchiveResult struct {
-	Archived []ArchiveItem
-	Metadata map[string]any
-}

internal/adapters/storage/bounded_read.go

@@ -0,0 +1,90 @@
+package storage
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"math"
+	"strings"
+)
+
+// ReadLimitError reports that a remote object exceeded its caller-owned read
+// limit. The limit is enforced against both available object metadata and the
+// bytes returned by the opened object body.
+type ReadLimitError struct {
+	Key      string
+	Limit    int64
+	Observed int64
+}
+
+func (e *ReadLimitError) Error() string {
+	return fmt.Sprintf("object %q exceeds %d-byte read limit (observed at least %d bytes)", e.Key, e.Limit, e.Observed)
+}
+
+// ReadObjectBounded opens one object version and retains at most maxBytes of
+// its content. Object metadata may reject an oversized body early, but a
+// limit-plus-one read always enforces the boundary when transfer begins.
+func ReadObjectBounded(ctx context.Context, store ObjectStore, key string, maxBytes int64) (info ObjectInfo, data []byte, err error) {
+	key = strings.TrimSpace(key)
+	if store == nil {
+		return ObjectInfo{}, nil, fmt.Errorf("read bounded object: store is required")
+	}
+	if key == "" {
+		return ObjectInfo{}, nil, fmt.Errorf("read bounded object: key is required")
+	}
+	if maxBytes <= 0 || maxBytes == math.MaxInt64 {
+		return ObjectInfo{}, nil, fmt.Errorf("read bounded object %q: limit must be between 1 and %d bytes", key, int64(math.MaxInt64-1))
+	}
+	if err := ctx.Err(); err != nil {
+		return ObjectInfo{}, nil, err
+	}
+
+	info, body, err := store.Read(ctx, key)
+	if err != nil {
+		return ObjectInfo{}, nil, err
+	}
+	if body == nil {
+		return ObjectInfo{}, nil, fmt.Errorf("read bounded object %q: store returned no body", key)
+	}
+	defer func() {
+		if closeErr := body.Close(); closeErr != nil {
+			data = nil
+			err = errors.Join(err, fmt.Errorf("close object %q: %w", key, closeErr))
+		}
+	}()
+
+	if info.Size > maxBytes {
+		return info, nil, &ReadLimitError{Key: key, Limit: maxBytes, Observed: info.Size}
+	}
+
+	data, err = io.ReadAll(io.LimitReader(contextReader{ctx: ctx, reader: body}, maxBytes+1))
+	if err != nil {
+		return info, nil, err
+	}
+	if err := ctx.Err(); err != nil {
+		return info, nil, err
+	}
+	if int64(len(data)) > maxBytes {
+		return info, nil, &ReadLimitError{Key: key, Limit: maxBytes, Observed: int64(len(data))}
+	}
+	return info, data, nil
+}
+
+type contextReader struct {
+	ctx    context.Context
+	reader io.Reader
+}
+
+func (r contextReader) Read(p []byte) (int, error) {
+	if err := r.ctx.Err(); err != nil {
+		return 0, err
+	}
+	n, err := r.reader.Read(p)
+	if err == nil {
+		if contextErr := r.ctx.Err(); contextErr != nil {
+			return n, contextErr
+		}
+	}
+	return n, err
+}

internal/adapters/storage/bounded_read_test.go

@@ -0,0 +1,135 @@
+package storage
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"io"
+	"testing"
+)
+
+func TestReadObjectBoundedAcceptsExactLimitWithAbsentSizeMetadata(t *testing.T) {
+	body := &trackingReadCloser{reader: bytes.NewReader([]byte("12345678")), chunkSize: 2}
+	store := &boundedReadStore{read: func(context.Context, string) (ObjectInfo, io.ReadCloser, error) {
+		return ObjectInfo{Key: "control.json", ETag: "generation"}, body, nil
+	}}
+
+	info, data, err := ReadObjectBounded(context.Background(), store, "control.json", 8)
+	if err != nil {
+		t.Fatalf("ReadObjectBounded() error = %v", err)
+	}
+	if string(data) != "12345678" || info.ETag != "generation" {
+		t.Fatalf("ReadObjectBounded() = (%#v, %q), want opened object metadata and bytes", info, data)
+	}
+	if !body.closed {
+		t.Fatal("object body was not closed")
+	}
+}
+
+func TestReadObjectBoundedRejectsLimitPlusOneDespiteMissingOrInaccurateMetadata(t *testing.T) {
+	tests := []struct {
+		name         string
+		metadataSize int64
+	}{
+		{name: "missing", metadataSize: 0},
+		{name: "inaccurate", metadataSize: 2},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			body := &trackingReadCloser{reader: bytes.NewReader([]byte("123456789")), chunkSize: 1}
+			store := &boundedReadStore{read: func(context.Context, string) (ObjectInfo, io.ReadCloser, error) {
+				return ObjectInfo{Key: "control.json", Size: test.metadataSize}, body, nil
+			}}
+
+			_, data, err := ReadObjectBounded(context.Background(), store, "control.json", 8)
+			var limitErr *ReadLimitError
+			if !errors.As(err, &limitErr) {
+				t.Fatalf("ReadObjectBounded() error = %v, want ReadLimitError", err)
+			}
+			if data != nil || body.bytesRead != 9 || !body.closed {
+				t.Fatalf("data=%q bytes read=%d closed=%t, want nil, 9, true", data, body.bytesRead, body.closed)
+			}
+		})
+	}
+}
+
+func TestReadObjectBoundedRejectsOversizedMetadataBeforeTransfer(t *testing.T) {
+	body := &trackingReadCloser{reader: bytes.NewReader([]byte("small"))}
+	store := &boundedReadStore{read: func(context.Context, string) (ObjectInfo, io.ReadCloser, error) {
+		return ObjectInfo{Key: "control.json", Size: 9}, body, nil
+	}}
+
+	_, _, err := ReadObjectBounded(context.Background(), store, "control.json", 8)
+	var limitErr *ReadLimitError
+	if !errors.As(err, &limitErr) {
+		t.Fatalf("ReadObjectBounded() error = %v, want ReadLimitError", err)
+	}
+	if body.bytesRead != 0 || !body.closed {
+		t.Fatalf("bytes read=%d closed=%t, want zero-byte transfer and closed body", body.bytesRead, body.closed)
+	}
+}
+
+func TestReadObjectBoundedPropagatesCancellationAndClosesBody(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	body := &trackingReadCloser{reader: bytes.NewReader([]byte("12345678")), chunkSize: 1, afterRead: cancel}
+	store := &boundedReadStore{read: func(context.Context, string) (ObjectInfo, io.ReadCloser, error) {
+		return ObjectInfo{Key: "control.json"}, body, nil
+	}}
+
+	_, data, err := ReadObjectBounded(ctx, store, "control.json", 8)
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("ReadObjectBounded() error = %v, want context cancellation", err)
+	}
+	if data != nil || body.bytesRead != 1 || !body.closed {
+		t.Fatalf("data=%q bytes read=%d closed=%t, want nil, 1, true", data, body.bytesRead, body.closed)
+	}
+}
+
+func TestReadObjectBoundedReturnsCloseFailure(t *testing.T) {
+	closeErr := errors.New("close failed")
+	body := &trackingReadCloser{reader: bytes.NewReader([]byte("ok")), closeErr: closeErr}
+	store := &boundedReadStore{read: func(context.Context, string) (ObjectInfo, io.ReadCloser, error) {
+		return ObjectInfo{Key: "control.json", Size: 2}, body, nil
+	}}
+
+	_, data, err := ReadObjectBounded(context.Background(), store, "control.json", 8)
+	if !errors.Is(err, closeErr) || data != nil || !body.closed {
+		t.Fatalf("data=%q error=%v closed=%t, want close failure and no retained data", data, err, body.closed)
+	}
+}
+
+type boundedReadStore struct {
+	ObjectStore
+	read func(context.Context, string) (ObjectInfo, io.ReadCloser, error)
+}
+
+func (s *boundedReadStore) Read(ctx context.Context, key string) (ObjectInfo, io.ReadCloser, error) {
+	return s.read(ctx, key)
+}
+
+type trackingReadCloser struct {
+	reader    io.Reader
+	chunkSize int
+	afterRead func()
+	closeErr  error
+	bytesRead int
+	closed    bool
+}
+
+func (r *trackingReadCloser) Read(p []byte) (int, error) {
+	if r.chunkSize > 0 && len(p) > r.chunkSize {
+		p = p[:r.chunkSize]
+	}
+	n, err := r.reader.Read(p)
+	r.bytesRead += n
+	if n > 0 && r.afterRead != nil {
+		r.afterRead()
+		r.afterRead = nil
+	}
+	return n, err
+}
+
+func (r *trackingReadCloser) Close() error {
+	r.closed = true
+	return r.closeErr
+}

internal/adapters/storage/download_writer.go

@@ -0,0 +1,23 @@
+package storage
+
+import (
+	"context"
+	"fmt"
+	"io"
+)
+
+// WriterDownloader is implemented by storage backends that stream an object
+// into a caller-owned file handle.
+type WriterDownloader interface {
+	DownloadTo(ctx context.Context, key string, destination io.Writer) error
+}
+
+// DownloadTo streams one object into destination. Destination-confined callers
+// require this capability rather than granting a backend a mutable pathname.
+func DownloadTo(ctx context.Context, store ObjectStore, key string, destination io.Writer) error {
+	writer, ok := store.(WriterDownloader)
+	if !ok {
+		return fmt.Errorf("object store does not support handle-confined downloads")
+	}
+	return writer.DownloadTo(ctx, key, destination)
+}

internal/adapters/storage/factory.go

@@ -0,0 +1,28 @@
+package storage
+
+import (
+	"context"
+	"fmt"
+	"strings"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/config"
+)
+
+// NewObjectStoreFromConfig constructs a remote object store from resolved config.
+func NewObjectStoreFromConfig(ctx context.Context, cfg *config.Config) (ObjectStore, error) {
+	if cfg == nil || cfg.Pipeline == nil {
+		return nil, fmt.Errorf("pipeline config is required")
+	}
+
+	switch strings.ToLower(strings.TrimSpace(cfg.Pipeline.Storage.Backend)) {
+	case config.StorageBackendS3:
+		if cfg.Pipeline.Storage.S3 == nil {
+			return nil, fmt.Errorf("pipeline.storage.s3 is required when pipeline.storage.backend is s3")
+		}
+		return NewS3BackendFromConfig(ctx, *cfg.Pipeline.Storage.S3)
+	case "", config.StorageBackendLocal:
+		return nil, fmt.Errorf("no remote object store backend is configured")
+	default:
+		return nil, fmt.Errorf("unsupported pipeline.storage.backend %q", cfg.Pipeline.Storage.Backend)
+	}
+}

internal/adapters/storage/factory_test.go

@@ -0,0 +1,88 @@
+package storage
+
+import (
+	"context"
+	"strings"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/config"
+)
+
+func TestNewObjectStoreFromConfigBuildsS3WhenBackendIsS3(t *testing.T) {
+	original := newS3Client
+	t.Cleanup(func() { newS3Client = original })
+	newS3Client = func(_ context.Context, _ s3ClientOptions) (s3API, error) {
+		return &fakeS3API{}, nil
+	}
+
+	store, err := NewObjectStoreFromConfig(context.Background(), &config.Config{
+		Pipeline: &config.PipelineConfig{
+			Storage: config.StorageConfig{
+				Backend: "s3",
+				S3: &config.StorageS3Config{
+					Bucket: "my-archive",
+				},
+			},
+		},
+	})
+	if err != nil {
+		t.Fatalf("NewObjectStoreFromConfig() error = %v", err)
+	}
+	if _, ok := store.(*S3Backend); !ok {
+		t.Fatalf("store type = %T, want *S3Backend", store)
+	}
+}
+
+func TestNewObjectStoreFromConfigRequiresS3ConfigWhenBackendIsS3(t *testing.T) {
+	_, err := NewObjectStoreFromConfig(context.Background(), &config.Config{
+		Pipeline: &config.PipelineConfig{
+			Storage: config.StorageConfig{Backend: "s3"},
+		},
+	})
+	if err == nil || !strings.Contains(err.Error(), "pipeline.storage.s3 is required") {
+		t.Fatalf("NewObjectStoreFromConfig() error = %v, want missing storage.s3 error", err)
+	}
+}
+
+func TestNewObjectStoreFromConfigNoRemoteBackendConfigured(t *testing.T) {
+	_, err := NewObjectStoreFromConfig(context.Background(), &config.Config{
+		Pipeline: &config.PipelineConfig{
+			Storage: config.StorageConfig{Backend: "local"},
+		},
+	})
+	if err == nil || !strings.Contains(err.Error(), "no remote object store backend is configured") {
+		t.Fatalf("NewObjectStoreFromConfig() error = %v, want no-backend error", err)
+	}
+}
+
+func TestNewObjectStoreFromConfigDoesNotInferS3FromProviderFields(t *testing.T) {
+	called := false
+	original := newS3Client
+	t.Cleanup(func() { newS3Client = original })
+	newS3Client = func(_ context.Context, _ s3ClientOptions) (s3API, error) {
+		called = true
+		return &fakeS3API{}, nil
+	}
+
+	_, err := NewObjectStoreFromConfig(context.Background(), &config.Config{
+		Pipeline: &config.PipelineConfig{Storage: config.StorageConfig{
+			Backend: config.StorageBackendLocal,
+			S3:      &config.StorageS3Config{Bucket: "my-archive"},
+		}},
+	})
+	if err == nil || !strings.Contains(err.Error(), "no remote object store backend is configured") {
+		t.Fatalf("NewObjectStoreFromConfig() error = %v, want no-backend error", err)
+	}
+	if called {
+		t.Fatal("NewObjectStoreFromConfig() constructed S3 from incidental provider fields")
+	}
+}
+
+func TestNewObjectStoreFromConfigRejectsUnknownBackend(t *testing.T) {
+	_, err := NewObjectStoreFromConfig(context.Background(), &config.Config{
+		Pipeline: &config.PipelineConfig{Storage: config.StorageConfig{Backend: "s33"}},
+	})
+	if err == nil || !strings.Contains(err.Error(), "unsupported pipeline.storage.backend") {
+		t.Fatalf("NewObjectStoreFromConfig() error = %v, want unsupported-backend error", err)
+	}
+}

internal/adapters/storage/fake.go

@@ -1,40 +1,341 @@
 package storage
 
-import "context"
+import (
+	"bytes"
+	"context"
+	"crypto/sha256"
+	"encoding/hex"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+	"sync"
+	"time"
+)
 
-// NoopBackend is a deterministic no-op archive/storage adapter.
-type NoopBackend struct{}
-
-// Archive returns the requested items as archived with placeholder metadata.
-func (n *NoopBackend) Archive(ctx context.Context, req ArchiveRequest) (ArchiveResult, error) {
-	if err := ctx.Err(); err != nil {
-		return ArchiveResult{}, err
-	}
-	return ArchiveResult{Archived: append([]ArchiveItem(nil), req.Items...), Metadata: map[string]any{"placeholder": true}}, nil
-}
-
-// FakeBackend captures archive requests and returns deterministic responses.
+// FakeBackend provides a deterministic in-memory object store for tests.
 type FakeBackend struct {
-	Requests []ArchiveRequest
-	Err      error
-	Result   ArchiveResult
+	mu sync.RWMutex
+
+	Objects   map[string]FakeObject
+	Uploads   []FakeUploadCall
+	Downloads []FakeDownloadCall
+	Reads     []FakeReadCall
+
+	ListErr      error
+	DownloadErr  error
+	UploadErr    error
+	ExistsErr    error
+	UploadHook   func(FakeUploadCall) error
+	DownloadHook func(FakeDownloadCall) error
 }
 
-// Archive records request and returns configured response.
-func (f *FakeBackend) Archive(ctx context.Context, req ArchiveRequest) (ArchiveResult, error) {
-	if err := ctx.Err(); err != nil {
-		return ArchiveResult{}, err
-	}
-	f.Requests = append(f.Requests, req)
-	if f.Err != nil {
-		return ArchiveResult{}, f.Err
-	}
-	res := f.Result
-	if res.Archived == nil {
-		res.Archived = append([]ArchiveItem(nil), req.Items...)
-	}
-	if res.Metadata == nil {
-		res.Metadata = map[string]any{"fake": true}
-	}
-	return res, nil
+// FakeUploadCall captures one upload invocation in call order.
+type FakeUploadCall struct {
+	LocalPath string
+	Key       string
+	Options   UploadOptions
+}
+
+// FakeDownloadCall captures one download invocation in call order.
+type FakeDownloadCall struct {
+	Key       string
+	LocalPath string
+	Bytes     int64
+}
+
+// FakeReadCall captures one opened object in call order.
+type FakeReadCall struct {
+	Key string
+}
+
+// FakeObject is a deterministic fake object-store record.
+type FakeObject struct {
+	Key          string
+	Data         []byte
+	Metadata     map[string]string
+	ETag         string
+	LastModified *time.Time
+}
+
+// SeedObject inserts or replaces an object in the fake object store.
+func (f *FakeBackend) SeedObject(obj FakeObject) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	f.seedObject(obj)
+}
+
+func (f *FakeBackend) seedObject(obj FakeObject) {
+	if f.Objects == nil {
+		f.Objects = map[string]FakeObject{}
+	}
+	key := normalizeObjectKey(obj.Key)
+	obj.Key = key
+	obj.Data = append([]byte(nil), obj.Data...)
+	obj.Metadata = copyMetadata(obj.Metadata)
+	if obj.ETag == "" {
+		obj.ETag = fakeObjectETag(obj.Data)
+	}
+	f.Objects[key] = obj
+}
+
+// List returns deterministic prefix-filtered objects.
+func (f *FakeBackend) List(ctx context.Context, prefix string) ([]ObjectInfo, error) {
+	if err := ctx.Err(); err != nil {
+		return nil, err
+	}
+	if f.ListErr != nil {
+		return nil, f.ListErr
+	}
+
+	f.mu.RLock()
+	defer f.mu.RUnlock()
+	normalizedPrefix := normalizeObjectKey(prefix)
+	keys := make([]string, 0, len(f.Objects))
+	for key := range f.Objects {
+		if strings.HasPrefix(key, normalizedPrefix) {
+			keys = append(keys, key)
+		}
+	}
+	sort.Strings(keys)
+
+	out := make([]ObjectInfo, 0, len(keys))
+	for _, key := range keys {
+		obj := f.Objects[key]
+		out = append(out, ObjectInfo{
+			Key:          obj.Key,
+			Size:         int64(len(obj.Data)),
+			ETag:         obj.ETag,
+			LastModified: obj.LastModified,
+		})
+	}
+	return out, nil
+}
+
+// Read returns a stable object body and the generation observed with it.
+func (f *FakeBackend) Read(ctx context.Context, key string) (ObjectInfo, io.ReadCloser, error) {
+	if err := ctx.Err(); err != nil {
+		return ObjectInfo{}, nil, err
+	}
+	if f.DownloadErr != nil {
+		return ObjectInfo{}, nil, f.DownloadErr
+	}
+	normalizedKey := normalizeObjectKey(key)
+	f.mu.RLock()
+	obj, ok := f.Objects[normalizedKey]
+	if ok {
+		obj.Data = append([]byte(nil), obj.Data...)
+		obj.Metadata = copyMetadata(obj.Metadata)
+	}
+	f.mu.RUnlock()
+	if !ok {
+		return ObjectInfo{}, nil, fmt.Errorf("read object %q: %w", normalizedKey, os.ErrNotExist)
+	}
+	f.mu.Lock()
+	f.Reads = append(f.Reads, FakeReadCall{Key: normalizedKey})
+	f.mu.Unlock()
+	return ObjectInfo{Key: obj.Key, Size: int64(len(obj.Data)), ETag: obj.ETag, LastModified: obj.LastModified}, io.NopCloser(bytes.NewReader(obj.Data)), nil
+}
+
+// DownloadTo writes one object to a caller-owned destination writer.
+func (f *FakeBackend) DownloadTo(ctx context.Context, key string, destination io.Writer) error {
+	if err := ctx.Err(); err != nil {
+		return err
+	}
+	if f.DownloadErr != nil {
+		return f.DownloadErr
+	}
+	if destination == nil {
+		return fmt.Errorf("download object: destination writer is required")
+	}
+	if f.DownloadHook != nil {
+		if err := f.DownloadHook(FakeDownloadCall{Key: normalizeObjectKey(key)}); err != nil {
+			return err
+		}
+	}
+
+	_, source, err := f.Read(ctx, key)
+	if err != nil {
+		return err
+	}
+	defer source.Close()
+	count, err := io.Copy(destination, source)
+	if err != nil {
+		return fmt.Errorf("download object %q: write destination: %w", key, err)
+	}
+	f.mu.Lock()
+	f.Downloads = append(f.Downloads, FakeDownloadCall{Key: normalizeObjectKey(key), Bytes: count})
+	f.mu.Unlock()
+	return nil
+}
+
+// Download writes one object to a local path.
+func (f *FakeBackend) Download(ctx context.Context, key, localPath string) error {
+	if strings.TrimSpace(localPath) == "" {
+		return fmt.Errorf("download object: local path is required")
+	}
+	if err := os.MkdirAll(filepath.Dir(localPath), 0o755); err != nil {
+		return fmt.Errorf("download object %q: create parent directory: %w", key, err)
+	}
+	destination, err := os.Create(localPath)
+	if err != nil {
+		return fmt.Errorf("download object %q: create local file: %w", key, err)
+	}
+	defer destination.Close()
+	return f.DownloadTo(ctx, key, destination)
+}
+
+// Upload reads a local file and stores it under key.
+func (f *FakeBackend) Upload(ctx context.Context, localPath, key string, opts UploadOptions) (ObjectInfo, error) {
+	if err := ctx.Err(); err != nil {
+		return ObjectInfo{}, err
+	}
+	if f.UploadErr != nil {
+		return ObjectInfo{}, f.UploadErr
+	}
+	if strings.TrimSpace(localPath) == "" {
+		return ObjectInfo{}, fmt.Errorf("upload object: local path is required")
+	}
+	if strings.TrimSpace(key) == "" {
+		return ObjectInfo{}, fmt.Errorf("upload object: key is required")
+	}
+
+	file, err := os.Open(localPath)
+	if err != nil {
+		return ObjectInfo{}, fmt.Errorf("upload object %q from %q: %w", key, localPath, err)
+	}
+	defer file.Close()
+	return f.uploadReader(ctx, file, key, opts, localPath)
+}
+
+// UploadReader stores content provided by a caller-owned reader.
+func (f *FakeBackend) UploadReader(ctx context.Context, source io.Reader, key string, opts UploadOptions) (ObjectInfo, error) {
+	return f.uploadReader(ctx, source, key, opts, "reader")
+}
+
+func (f *FakeBackend) uploadReader(ctx context.Context, source io.Reader, key string, opts UploadOptions, localPath string) (ObjectInfo, error) {
+	if err := ctx.Err(); err != nil {
+		return ObjectInfo{}, err
+	}
+	if f.UploadErr != nil {
+		return ObjectInfo{}, f.UploadErr
+	}
+	if source == nil {
+		return ObjectInfo{}, fmt.Errorf("upload object: source is required")
+	}
+	if strings.TrimSpace(key) == "" {
+		return ObjectInfo{}, fmt.Errorf("upload object: key is required")
+	}
+
+	data, err := io.ReadAll(source)
+	if err != nil {
+		return ObjectInfo{}, fmt.Errorf("upload object %q from %q: %w", key, localPath, err)
+	}
+
+	normalizedKey := normalizeObjectKey(key)
+	call := FakeUploadCall{
+		LocalPath: localPath,
+		Key:       normalizedKey,
+		Options: UploadOptions{
+			Metadata:    copyMetadata(opts.Metadata),
+			ContentType: opts.ContentType,
+		},
+	}
+	f.mu.Lock()
+	f.Uploads = append(f.Uploads, call)
+	f.mu.Unlock()
+	if f.UploadHook != nil {
+		if err := f.UploadHook(call); err != nil {
+			return ObjectInfo{}, err
+		}
+	}
+	return f.storeUploadedObject(normalizedKey, data, opts), nil
+}
+
+// UploadConditional atomically checks and replaces one mutable object.
+func (f *FakeBackend) UploadConditional(ctx context.Context, source io.Reader, key string, opts UploadOptions, condition WriteCondition) (ObjectInfo, error) {
+	if err := ctx.Err(); err != nil {
+		return ObjectInfo{}, err
+	}
+	if err := validateWriteCondition(condition); err != nil {
+		return ObjectInfo{}, err
+	}
+	if f.UploadErr != nil {
+		return ObjectInfo{}, f.UploadErr
+	}
+	if source == nil {
+		return ObjectInfo{}, fmt.Errorf("upload object: source is required")
+	}
+	normalizedKey := normalizeObjectKey(key)
+	if normalizedKey == "" {
+		return ObjectInfo{}, fmt.Errorf("upload object: key is required")
+	}
+	data, err := io.ReadAll(source)
+	if err != nil {
+		return ObjectInfo{}, fmt.Errorf("upload object %q: %w", normalizedKey, err)
+	}
+	call := FakeUploadCall{Key: normalizedKey, Options: UploadOptions{Metadata: copyMetadata(opts.Metadata), ContentType: opts.ContentType}}
+	f.mu.Lock()
+	f.Uploads = append(f.Uploads, call)
+	f.mu.Unlock()
+	if f.UploadHook != nil {
+		if err := f.UploadHook(call); err != nil {
+			return ObjectInfo{}, err
+		}
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	existing, found := f.Objects[normalizedKey]
+	if condition.RequireAbsent && found {
+		return ObjectInfo{}, ErrConditionNotMet
+	}
+	if expected := strings.TrimSpace(condition.MatchETag); expected != "" && (!found || existing.ETag != expected) {
+		return ObjectInfo{}, ErrConditionNotMet
+	}
+	return f.storeUploadedObjectLocked(normalizedKey, data, opts), nil
+}
+
+func (f *FakeBackend) storeUploadedObject(key string, data []byte, opts UploadOptions) ObjectInfo {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	return f.storeUploadedObjectLocked(key, data, opts)
+}
+
+func (f *FakeBackend) storeUploadedObjectLocked(key string, data []byte, opts UploadOptions) ObjectInfo {
+	now := time.Now().UTC()
+	obj := FakeObject{Key: key, Data: append([]byte(nil), data...), Metadata: copyMetadata(opts.Metadata), LastModified: &now}
+	f.seedObject(obj)
+	return ObjectInfo{Key: key, Size: int64(len(data)), ETag: fakeObjectETag(data), LastModified: &now}
+}
+
+func fakeObjectETag(data []byte) string {
+	sum := sha256.Sum256(data)
+	return hex.EncodeToString(sum[:])
+}
+
+// Exists checks object presence.
+func (f *FakeBackend) Exists(ctx context.Context, key string) (bool, error) {
+	if err := ctx.Err(); err != nil {
+		return false, err
+	}
+	if f.ExistsErr != nil {
+		return false, f.ExistsErr
+	}
+	f.mu.RLock()
+	_, ok := f.Objects[normalizeObjectKey(key)]
+	f.mu.RUnlock()
+	return ok, nil
+}
+
+func copyMetadata(in map[string]string) map[string]string {
+	if len(in) == 0 {
+		return nil
+	}
+	out := make(map[string]string, len(in))
+	for k, v := range in {
+		out[k] = v
+	}
+	return out
 }

internal/adapters/storage/fake_test.go

@@ -3,29 +3,104 @@ package storage
 import (
 	"context"
 	"errors"
+	"os"
+	"path/filepath"
+	"strings"
 	"testing"
 )
 
-func TestFakeBackendCapturesRequestAndReturnsItems(t *testing.T) {
+func TestFakeBackendListPrefixFiltering(t *testing.T) {
 	fake := &FakeBackend{}
-	req := ArchiveRequest{SessionID: "s1", Items: []ArchiveItem{{Kind: "artifact", LocalPath: "artifacts/log.md"}}}
+	fake.SeedObject(FakeObject{Key: "dnd/campaigns/forsaken/audio/a.flac", Data: []byte("a")})
+	fake.SeedObject(FakeObject{Key: "dnd/campaigns/forsaken/audio/b.flac", Data: []byte("b")})
+	fake.SeedObject(FakeObject{Key: "dnd/campaigns/other/audio/c.flac", Data: []byte("c")})
 
-	res, err := fake.Archive(context.Background(), req)
+	items, err := fake.List(context.Background(), "dnd/campaigns/forsaken/audio/")
 	if err != nil {
-		t.Fatalf("Archive() error = %v", err)
+		t.Fatalf("List() error = %v", err)
 	}
-	if len(fake.Requests) != 1 || fake.Requests[0].SessionID != "s1" {
-		t.Fatalf("requests = %#v, want captured request", fake.Requests)
+	if len(items) != 2 {
+		t.Fatalf("List() len = %d, want 2", len(items))
 	}
-	if len(res.Archived) != 1 {
-		t.Fatalf("archived len = %d, want 1", len(res.Archived))
+	if items[0].Key != "dnd/campaigns/forsaken/audio/a.flac" || items[1].Key != "dnd/campaigns/forsaken/audio/b.flac" {
+		t.Fatalf("List() keys = %#v", items)
 	}
 }
 
-func TestFakeBackendError(t *testing.T) {
-	fake := &FakeBackend{Err: errors.New("boom")}
-	_, err := fake.Archive(context.Background(), ArchiveRequest{})
-	if err == nil {
-		t.Fatal("expected error, got nil")
+func TestFakeBackendDownload(t *testing.T) {
+	fake := &FakeBackend{}
+	fake.SeedObject(FakeObject{Key: "audio/a.flac", Data: []byte("audio-a")})
+
+	dst := filepath.Join(t.TempDir(), "nested", "a.flac")
+	if err := fake.Download(context.Background(), `audio\a.flac`, dst); err != nil {
+		t.Fatalf("Download() error = %v", err)
+	}
+	data, err := os.ReadFile(dst)
+	if err != nil {
+		t.Fatalf("ReadFile() error = %v", err)
+	}
+	if string(data) != "audio-a" {
+		t.Fatalf("downloaded content = %q, want %q", string(data), "audio-a")
+	}
+}
+
+func TestFakeBackendUploadAndExists(t *testing.T) {
+	fake := &FakeBackend{}
+
+	local := filepath.Join(t.TempDir(), "upload.txt")
+	if err := os.WriteFile(local, []byte("payload"), 0o644); err != nil {
+		t.Fatalf("WriteFile() error = %v", err)
+	}
+
+	info, err := fake.Upload(context.Background(), local, `runs\id\artifact.txt`, UploadOptions{
+		Metadata: map[string]string{"kind": "artifact"},
+	})
+	if err != nil {
+		t.Fatalf("Upload() error = %v", err)
+	}
+	if info.Key != "runs/id/artifact.txt" {
+		t.Fatalf("Upload() key = %q, want normalized key", info.Key)
+	}
+
+	ok, err := fake.Exists(context.Background(), "runs/id/artifact.txt")
+	if err != nil {
+		t.Fatalf("Exists() error = %v", err)
+	}
+	if !ok {
+		t.Fatal("Exists() = false, want true")
+	}
+}
+
+func TestFakeBackendConditionalUploadRejectsStaleGeneration(t *testing.T) {
+	fake := &FakeBackend{}
+	fake.SeedObject(FakeObject{Key: "locks.yml", Data: []byte("old")})
+	old := fake.Objects["locks.yml"].ETag
+	if _, err := fake.UploadConditional(context.Background(), strings.NewReader("new"), "locks.yml", UploadOptions{}, WriteCondition{MatchETag: old}); err != nil {
+		t.Fatalf("UploadConditional() error = %v", err)
+	}
+	if _, err := fake.UploadConditional(context.Background(), strings.NewReader("lost"), "locks.yml", UploadOptions{}, WriteCondition{MatchETag: old}); !errors.Is(err, ErrConditionNotMet) {
+		t.Fatalf("UploadConditional() error = %v, want ErrConditionNotMet", err)
+	}
+	if got := string(fake.Objects["locks.yml"].Data); got != "new" {
+		t.Fatalf("locks object = %q, want successful replacement preserved", got)
+	}
+}
+
+func TestFakeBackendObjectErrors(t *testing.T) {
+	fake := &FakeBackend{DownloadErr: errors.New("download fail"), UploadErr: errors.New("upload fail"), ListErr: errors.New("list fail"), ExistsErr: errors.New("exists fail")}
+
+	if _, err := fake.List(context.Background(), "x"); err == nil || !strings.Contains(err.Error(), "list fail") {
+		t.Fatalf("List() error = %v, want list fail", err)
+	}
+	if err := fake.Download(context.Background(), "x", filepath.Join(t.TempDir(), "x")); err == nil || !strings.Contains(err.Error(), "download fail") {
+		t.Fatalf("Download() error = %v, want download fail", err)
+	}
+	local := filepath.Join(t.TempDir(), "x.txt")
+	_ = os.WriteFile(local, []byte("x"), 0o644)
+	if _, err := fake.Upload(context.Background(), local, "x", UploadOptions{}); err == nil || !strings.Contains(err.Error(), "upload fail") {
+		t.Fatalf("Upload() error = %v, want upload fail", err)
+	}
+	if _, err := fake.Exists(context.Background(), "x"); err == nil || !strings.Contains(err.Error(), "exists fail") {
+		t.Fatalf("Exists() error = %v, want exists fail", err)
 	}
 }

internal/adapters/storage/keys.go

@@ -0,0 +1,8 @@
+package storage
+
+import "strings"
+
+func normalizeObjectKey(key string) string {
+	normalized := strings.ReplaceAll(strings.TrimSpace(key), "\\", "/")
+	return strings.TrimLeft(normalized, "/")
+}

internal/adapters/storage/keys_test.go

@@ -0,0 +1,21 @@
+package storage
+
+import "testing"
+
+func TestNormalizeObjectKey(t *testing.T) {
+	tests := []struct {
+		in   string
+		want string
+	}{
+		{in: `dnd\campaigns\forsaken\a.flac`, want: "dnd/campaigns/forsaken/a.flac"},
+		{in: " /dnd/campaigns/forsaken/a.flac ", want: "dnd/campaigns/forsaken/a.flac"},
+		{in: "//dnd/campaigns/forsaken/a.flac", want: "dnd/campaigns/forsaken/a.flac"},
+		{in: "", want: ""},
+	}
+
+	for _, tt := range tests {
+		if got := normalizeObjectKey(tt.in); got != tt.want {
+			t.Fatalf("normalizeObjectKey(%q) = %q, want %q", tt.in, got, tt.want)
+		}
+	}
+}

internal/adapters/storage/object_store.go

@@ -0,0 +1,53 @@
+package storage
+
+import (
+	"context"
+	"errors"
+	"io"
+	"time"
+)
+
+// ErrConditionNotMet reports that an object changed or already existed before a
+// conditional write could be committed.
+var ErrConditionNotMet = errors.New("object write condition not met")
+
+// ReaderUploader streams caller-owned, already-opened content to object storage.
+// Callers retain source-selection and filesystem-confinement policy.
+type ReaderUploader interface {
+	UploadReader(ctx context.Context, source io.Reader, key string, opts UploadOptions) (ObjectInfo, error)
+}
+
+// ObjectStore is a remote object storage boundary used by prepare, restore, and publish work.
+//
+// Key invariant:
+// callers pass full bucket-relative object keys. Backend implementations do not
+// infer Narratio session semantics and do not prepend root prefixes.
+type ObjectStore interface {
+	List(ctx context.Context, prefix string) ([]ObjectInfo, error)
+	Read(ctx context.Context, key string) (ObjectInfo, io.ReadCloser, error)
+	Download(ctx context.Context, key, localPath string) error
+	Upload(ctx context.Context, localPath, key string, opts UploadOptions) (ObjectInfo, error)
+	UploadConditional(ctx context.Context, source io.Reader, key string, opts UploadOptions, condition WriteCondition) (ObjectInfo, error)
+	Exists(ctx context.Context, key string) (bool, error)
+}
+
+// ObjectInfo describes one object in remote storage.
+type ObjectInfo struct {
+	Key          string
+	Size         int64
+	ETag         string
+	LastModified *time.Time
+}
+
+// UploadOptions configures optional object upload metadata.
+type UploadOptions struct {
+	Metadata    map[string]string
+	ContentType string
+}
+
+// WriteCondition protects a mutable object update against a stale snapshot.
+// Exactly one condition is required by UploadConditional.
+type WriteCondition struct {
+	MatchETag     string
+	RequireAbsent bool
+}

internal/adapters/storage/s3_backend.go

@@ -0,0 +1,372 @@
+package storage
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
+	"time"
+
+	awsconfig "github.com/aws/aws-sdk-go-v2/config"
+	"github.com/aws/aws-sdk-go-v2/credentials"
+	"github.com/aws/aws-sdk-go-v2/service/s3"
+	"github.com/aws/aws-sdk-go-v2/service/s3/types"
+	"github.com/aws/smithy-go"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/config"
+)
+
+type s3API interface {
+	ListObjectsV2(ctx context.Context, params *s3.ListObjectsV2Input, optFns ...func(*s3.Options)) (*s3.ListObjectsV2Output, error)
+	GetObject(ctx context.Context, params *s3.GetObjectInput, optFns ...func(*s3.Options)) (*s3.GetObjectOutput, error)
+	PutObject(ctx context.Context, params *s3.PutObjectInput, optFns ...func(*s3.Options)) (*s3.PutObjectOutput, error)
+	HeadObject(ctx context.Context, params *s3.HeadObjectInput, optFns ...func(*s3.Options)) (*s3.HeadObjectOutput, error)
+}
+
+// S3Backend is an ObjectStore implementation backed by S3-compatible APIs.
+type S3Backend struct {
+	bucket string
+	client s3API
+}
+
+type s3ClientOptions struct {
+	Region         string
+	Endpoint       string
+	ForcePathStyle bool
+	AccessKeyID    string
+	SecretKey      string
+}
+
+var newS3Client = func(ctx context.Context, opts s3ClientOptions) (s3API, error) {
+	loadOpts := make([]func(*awsconfig.LoadOptions) error, 0, 1)
+	if strings.TrimSpace(opts.Region) != "" {
+		loadOpts = append(loadOpts, awsconfig.WithRegion(strings.TrimSpace(opts.Region)))
+	}
+	if strings.TrimSpace(opts.AccessKeyID) != "" && strings.TrimSpace(opts.SecretKey) != "" {
+		loadOpts = append(loadOpts, awsconfig.WithCredentialsProvider(
+			credentials.NewStaticCredentialsProvider(
+				strings.TrimSpace(opts.AccessKeyID),
+				strings.TrimSpace(opts.SecretKey),
+				"",
+			),
+		))
+	}
+	awsCfg, err := awsconfig.LoadDefaultConfig(ctx, loadOpts...)
+	if err != nil {
+		return nil, fmt.Errorf("load aws config: %w", err)
+	}
+
+	return s3.NewFromConfig(awsCfg, func(o *s3.Options) {
+		if strings.TrimSpace(opts.Endpoint) != "" {
+			endpoint := strings.TrimSpace(opts.Endpoint)
+			o.BaseEndpoint = &endpoint
+		}
+		o.UsePathStyle = opts.ForcePathStyle
+	}), nil
+}
+
+// NewS3BackendFromConfig builds an S3 backend from resolved config.
+func NewS3BackendFromConfig(ctx context.Context, cfg config.StorageS3Config) (*S3Backend, error) {
+	bucket := strings.TrimSpace(cfg.Bucket)
+	if bucket == "" {
+		return nil, fmt.Errorf("storage.s3.bucket is required")
+	}
+
+	client, err := newS3Client(ctx, s3ClientOptions{
+		Region:         cfg.Region,
+		Endpoint:       cfg.Endpoint,
+		ForcePathStyle: cfg.ForcePathStyle,
+		AccessKeyID:    s3CredentialFromEnv(orDefaultEnvName(cfg.AccessKeyIDEnv, config.DefaultS3AccessKeyIDEnv)),
+		SecretKey:      s3CredentialFromEnv(orDefaultEnvName(cfg.SecretKeyEnv, config.DefaultS3SecretAccessKeyEnv)),
+	})
+	if err != nil {
+		return nil, fmt.Errorf("build s3 client: %w", err)
+	}
+
+	return &S3Backend{
+		bucket: bucket,
+		client: client,
+	}, nil
+}
+
+func s3CredentialFromEnv(envVarName string) string {
+	name := strings.TrimSpace(envVarName)
+	if name == "" {
+		return ""
+	}
+	value, ok := os.LookupEnv(name)
+	if !ok {
+		return ""
+	}
+	return strings.TrimSpace(value)
+}
+
+func orDefaultEnvName(name, fallback string) string {
+	trimmed := strings.TrimSpace(name)
+	if trimmed == "" {
+		return fallback
+	}
+	return trimmed
+}
+
+// List returns objects under prefix.
+func (b *S3Backend) List(ctx context.Context, prefix string) ([]ObjectInfo, error) {
+	normalizedPrefix := normalizeObjectKey(prefix)
+	out := make([]ObjectInfo, 0)
+	var token *string
+	seenTokens := map[string]struct{}{}
+
+	for {
+		resp, err := b.client.ListObjectsV2(ctx, &s3.ListObjectsV2Input{
+			Bucket:            &b.bucket,
+			Prefix:            &normalizedPrefix,
+			ContinuationToken: token,
+		})
+		if err != nil {
+			return nil, fmt.Errorf("list objects under %q: %w", normalizedPrefix, err)
+		}
+
+		for _, item := range resp.Contents {
+			var lastModified *time.Time
+			if item.LastModified != nil {
+				t := *item.LastModified
+				lastModified = &t
+			}
+			out = append(out, ObjectInfo{
+				Key:          normalizeObjectKey(valueOrEmpty(item.Key)),
+				Size:         valueOrZeroInt64(item.Size),
+				ETag:         strings.Trim(valueOrEmpty(item.ETag), "\""),
+				LastModified: lastModified,
+			})
+		}
+
+		if !valueOrFalseBool(resp.IsTruncated) {
+			break
+		}
+		next := strings.TrimSpace(valueOrEmpty(resp.NextContinuationToken))
+		if next == "" {
+			return nil, fmt.Errorf("s3 list objects bucket %q prefix %q: truncated response has an empty continuation token", b.bucket, normalizedPrefix)
+		}
+		if _, repeated := seenTokens[next]; repeated {
+			return nil, fmt.Errorf("s3 list objects bucket %q prefix %q: truncated response repeated continuation token", b.bucket, normalizedPrefix)
+		}
+		seenTokens[next] = struct{}{}
+		token = &next
+	}
+	return out, nil
+}
+
+// Read retrieves an object together with the generation observed for its body.
+func (b *S3Backend) Read(ctx context.Context, key string) (ObjectInfo, io.ReadCloser, error) {
+	normalizedKey := normalizeObjectKey(key)
+	resp, err := b.client.GetObject(ctx, &s3.GetObjectInput{Bucket: &b.bucket, Key: &normalizedKey})
+	if err != nil {
+		if isS3NotFound(err) {
+			return ObjectInfo{}, nil, fmt.Errorf("read object %q: %w", normalizedKey, os.ErrNotExist)
+		}
+		return ObjectInfo{}, nil, fmt.Errorf("read object %q: %w", normalizedKey, err)
+	}
+	var lastModified *time.Time
+	if resp.LastModified != nil {
+		t := *resp.LastModified
+		lastModified = &t
+	}
+	return ObjectInfo{
+		Key: normalizedKey, Size: valueOrZeroInt64(resp.ContentLength),
+		ETag: strings.Trim(valueOrEmpty(resp.ETag), "\""), LastModified: lastModified,
+	}, resp.Body, nil
+}
+
+// DownloadTo retrieves one object into the caller-owned destination writer.
+func (b *S3Backend) DownloadTo(ctx context.Context, key string, destination io.Writer) error {
+	if destination == nil {
+		return fmt.Errorf("download object: destination writer is required")
+	}
+	_, body, err := b.Read(ctx, key)
+	if err != nil {
+		return fmt.Errorf("download object %q: %w", normalizeObjectKey(key), err)
+	}
+	defer body.Close()
+
+	if _, err := io.Copy(destination, body); err != nil {
+		return fmt.Errorf("download object %q: copy body: %w", normalizeObjectKey(key), err)
+	}
+	return nil
+}
+
+// Download retrieves one object to localPath, creating parent directories as needed.
+func (b *S3Backend) Download(ctx context.Context, key, localPath string) error {
+	if strings.TrimSpace(localPath) == "" {
+		return fmt.Errorf("download object: local path is required")
+	}
+	if err := os.MkdirAll(filepath.Dir(localPath), 0o755); err != nil {
+		return fmt.Errorf("download object %q: create parent directory: %w", key, err)
+	}
+	dst, err := os.Create(localPath)
+	if err != nil {
+		return fmt.Errorf("download object %q: create local file: %w", key, err)
+	}
+	defer dst.Close()
+	if err := b.DownloadTo(ctx, key, dst); err != nil {
+		return err
+	}
+	if err := dst.Sync(); err != nil {
+		return fmt.Errorf("download object %q: sync local file: %w", key, err)
+	}
+	return nil
+}
+
+// Upload sends a local file to key.
+func (b *S3Backend) Upload(ctx context.Context, localPath, key string, opts UploadOptions) (ObjectInfo, error) {
+	normalizedKey := normalizeObjectKey(key)
+	if strings.TrimSpace(localPath) == "" {
+		return ObjectInfo{}, fmt.Errorf("upload object: local path is required")
+	}
+	if normalizedKey == "" {
+		return ObjectInfo{}, fmt.Errorf("upload object: key is required")
+	}
+
+	file, err := os.Open(localPath)
+	if err != nil {
+		return ObjectInfo{}, fmt.Errorf("upload object %q from %q: %w", normalizedKey, localPath, err)
+	}
+	defer file.Close()
+
+	stat, err := file.Stat()
+	if err != nil {
+		return ObjectInfo{}, fmt.Errorf("upload object %q from %q: stat local file: %w", normalizedKey, localPath, err)
+	}
+	return b.uploadReader(ctx, file, key, opts, stat.Size(), WriteCondition{})
+}
+
+// UploadReader sends caller-owned content to key.
+func (b *S3Backend) UploadReader(ctx context.Context, source io.Reader, key string, opts UploadOptions) (ObjectInfo, error) {
+	return b.uploadReader(ctx, source, key, opts, 0, WriteCondition{})
+}
+
+// UploadConditional uploads a mutable object only when its observed generation
+// still matches, or when no object exists yet.
+func (b *S3Backend) UploadConditional(ctx context.Context, source io.Reader, key string, opts UploadOptions, condition WriteCondition) (ObjectInfo, error) {
+	if err := validateWriteCondition(condition); err != nil {
+		return ObjectInfo{}, err
+	}
+	return b.uploadReader(ctx, source, key, opts, 0, condition)
+}
+
+func (b *S3Backend) uploadReader(ctx context.Context, source io.Reader, key string, opts UploadOptions, size int64, condition WriteCondition) (ObjectInfo, error) {
+	normalizedKey := normalizeObjectKey(key)
+	if source == nil {
+		return ObjectInfo{}, fmt.Errorf("upload object: source is required")
+	}
+	if normalizedKey == "" {
+		return ObjectInfo{}, fmt.Errorf("upload object: key is required")
+	}
+
+	input := &s3.PutObjectInput{
+		Bucket:   &b.bucket,
+		Key:      &normalizedKey,
+		Body:     source,
+		Metadata: copyMetadata(opts.Metadata),
+	}
+	if strings.TrimSpace(opts.ContentType) != "" {
+		ct := strings.TrimSpace(opts.ContentType)
+		input.ContentType = &ct
+	}
+	if condition.RequireAbsent {
+		wildcard := "*"
+		input.IfNoneMatch = &wildcard
+	} else if expected := strings.TrimSpace(condition.MatchETag); expected != "" {
+		input.IfMatch = &expected
+	}
+
+	resp, err := b.client.PutObject(ctx, input)
+	if err != nil {
+		if isS3ConditionalConflict(err) {
+			return ObjectInfo{}, fmt.Errorf("upload object %q: %w", normalizedKey, ErrConditionNotMet)
+		}
+		return ObjectInfo{}, fmt.Errorf("upload object %q: %w", normalizedKey, err)
+	}
+
+	info := ObjectInfo{
+		Key:  normalizedKey,
+		ETag: strings.Trim(valueOrEmpty(resp.ETag), "\""),
+	}
+	if size > 0 {
+		info.Size = size
+	}
+	return info, nil
+}
+
+// Exists checks whether one object key exists.
+func (b *S3Backend) Exists(ctx context.Context, key string) (bool, error) {
+	normalizedKey := normalizeObjectKey(key)
+	_, err := b.client.HeadObject(ctx, &s3.HeadObjectInput{
+		Bucket: &b.bucket,
+		Key:    &normalizedKey,
+	})
+	if err == nil {
+		return true, nil
+	}
+
+	if isS3NotFound(err) {
+		return false, nil
+	}
+	return false, fmt.Errorf("head object %q: %w", normalizedKey, err)
+}
+
+func validateWriteCondition(condition WriteCondition) error {
+	if condition.RequireAbsent == (strings.TrimSpace(condition.MatchETag) != "") {
+		return fmt.Errorf("conditional upload requires exactly one of MatchETag or RequireAbsent")
+	}
+	return nil
+}
+
+func isS3NotFound(err error) bool {
+	var notFound *types.NotFound
+	if errors.As(err, &notFound) {
+		return true
+	}
+	var apiErr smithy.APIError
+	if errors.As(err, &apiErr) {
+		switch apiErr.ErrorCode() {
+		case "NotFound", "NoSuchKey", "404":
+			return true
+		}
+	}
+	return false
+}
+
+func isS3ConditionalConflict(err error) bool {
+	var apiErr smithy.APIError
+	if errors.As(err, &apiErr) {
+		switch apiErr.ErrorCode() {
+		case "PreconditionFailed", "ConditionalRequestConflict", "412", "409":
+			return true
+		}
+	}
+	return false
+}
+
+func valueOrEmpty(v *string) string {
+	if v == nil {
+		return ""
+	}
+	return *v
+}
+
+func valueOrZeroInt64(v *int64) int64 {
+	if v == nil {
+		return 0
+	}
+	return *v
+}
+
+func valueOrFalseBool(v *bool) bool {
+	if v == nil {
+		return false
+	}
+	return *v
+}

internal/adapters/storage/s3_backend_test.go

@@ -0,0 +1,345 @@
+package storage
+
+import (
+	"context"
+	"errors"
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/aws/aws-sdk-go-v2/service/s3"
+	"github.com/aws/aws-sdk-go-v2/service/s3/types"
+	"github.com/aws/smithy-go"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/config"
+)
+
+type fakeS3API struct {
+	listOut     *s3.ListObjectsV2Output
+	listOutputs []*s3.ListObjectsV2Output
+	listErr     error
+	listCalls   int
+
+	getBody         io.ReadCloser
+	getErr          error
+	getSize         *int64
+	getETag         *string
+	getLastModified *time.Time
+
+	putOut *s3.PutObjectOutput
+	putErr error
+
+	headErr error
+
+	lastList  *s3.ListObjectsV2Input
+	lastLists []*s3.ListObjectsV2Input
+	lastGet   *s3.GetObjectInput
+	lastPut   *s3.PutObjectInput
+	lastHead  *s3.HeadObjectInput
+}
+
+func (f *fakeS3API) ListObjectsV2(_ context.Context, params *s3.ListObjectsV2Input, _ ...func(*s3.Options)) (*s3.ListObjectsV2Output, error) {
+	f.lastList = params
+	f.lastLists = append(f.lastLists, params)
+	if f.listErr != nil {
+		return nil, f.listErr
+	}
+	if f.listCalls < len(f.listOutputs) {
+		out := f.listOutputs[f.listCalls]
+		f.listCalls++
+		return out, nil
+	}
+	if f.listOut == nil {
+		return &s3.ListObjectsV2Output{}, nil
+	}
+	return f.listOut, nil
+}
+
+func TestS3BackendListPaginatesAndRejectsNonProgressingTokens(t *testing.T) {
+	t.Run("multiple pages", func(t *testing.T) {
+		client := &fakeS3API{listOutputs: []*s3.ListObjectsV2Output{
+			{Contents: []types.Object{{Key: strPtr("prefix/a"), Size: int64Ptr(1)}}, IsTruncated: boolPtr(true), NextContinuationToken: strPtr("next")},
+			{Contents: []types.Object{{Key: strPtr("prefix/b"), Size: int64Ptr(2)}}, IsTruncated: boolPtr(false)},
+		}}
+		items, err := (&S3Backend{bucket: "bucket-1", client: client}).List(context.Background(), "prefix/")
+		if err != nil {
+			t.Fatalf("List() error = %v", err)
+		}
+		if len(items) != 2 || items[0].Key != "prefix/a" || items[1].Key != "prefix/b" {
+			t.Fatalf("List() items = %#v", items)
+		}
+		if len(client.lastLists) != 2 || client.lastLists[1].ContinuationToken == nil || *client.lastLists[1].ContinuationToken != "next" {
+			t.Fatalf("continuation calls = %#v", client.lastLists)
+		}
+	})
+
+	for _, test := range []struct {
+		name    string
+		outputs []*s3.ListObjectsV2Output
+		want    string
+	}{
+		{name: "empty", outputs: []*s3.ListObjectsV2Output{{IsTruncated: boolPtr(true)}}, want: "empty continuation token"},
+		{name: "repeated", outputs: []*s3.ListObjectsV2Output{{IsTruncated: boolPtr(true), NextContinuationToken: strPtr("again")}, {IsTruncated: boolPtr(true), NextContinuationToken: strPtr("again")}}, want: "repeated continuation token"},
+	} {
+		t.Run(test.name, func(t *testing.T) {
+			_, err := (&S3Backend{bucket: "bucket-1", client: &fakeS3API{listOutputs: test.outputs}}).List(context.Background(), "prefix/")
+			if err == nil || !strings.Contains(err.Error(), test.want) || !strings.Contains(err.Error(), "bucket-1") || !strings.Contains(err.Error(), "prefix/") {
+				t.Fatalf("List() error = %v, want contextual %q", err, test.want)
+			}
+		})
+	}
+}
+
+func (f *fakeS3API) GetObject(_ context.Context, params *s3.GetObjectInput, _ ...func(*s3.Options)) (*s3.GetObjectOutput, error) {
+	f.lastGet = params
+	if f.getErr != nil {
+		return nil, f.getErr
+	}
+	body := f.getBody
+	if body == nil {
+		body = io.NopCloser(strings.NewReader(""))
+	}
+	return &s3.GetObjectOutput{Body: body, ContentLength: f.getSize, ETag: f.getETag, LastModified: f.getLastModified}, nil
+}
+
+func (f *fakeS3API) PutObject(_ context.Context, params *s3.PutObjectInput, _ ...func(*s3.Options)) (*s3.PutObjectOutput, error) {
+	f.lastPut = params
+	if f.putErr != nil {
+		return nil, f.putErr
+	}
+	if f.putOut == nil {
+		return &s3.PutObjectOutput{}, nil
+	}
+	return f.putOut, nil
+}
+
+func (f *fakeS3API) HeadObject(_ context.Context, params *s3.HeadObjectInput, _ ...func(*s3.Options)) (*s3.HeadObjectOutput, error) {
+	f.lastHead = params
+	if f.headErr != nil {
+		return nil, f.headErr
+	}
+	return &s3.HeadObjectOutput{}, nil
+}
+
+func TestS3BackendListAndKeyNormalization(t *testing.T) {
+	lastModified := time.Date(2026, 5, 16, 12, 0, 0, 0, time.UTC)
+	client := &fakeS3API{
+		listOut: &s3.ListObjectsV2Output{
+			Contents: []types.Object{
+				{Key: strPtr(`dnd\campaigns\forsaken\a.flac`), Size: int64Ptr(7), ETag: strPtr(`"abc"`), LastModified: &lastModified},
+			},
+		},
+	}
+	backend := &S3Backend{bucket: "bucket-1", client: client}
+
+	items, err := backend.List(context.Background(), `dnd\campaigns\`)
+	if err != nil {
+		t.Fatalf("List() error = %v", err)
+	}
+	if len(items) != 1 {
+		t.Fatalf("List() len = %d, want 1", len(items))
+	}
+	if items[0].Key != "dnd/campaigns/forsaken/a.flac" {
+		t.Fatalf("List() key = %q, want normalized slash key", items[0].Key)
+	}
+	if items[0].ETag != "abc" {
+		t.Fatalf("List() ETag = %q, want %q", items[0].ETag, "abc")
+	}
+	if client.lastList == nil || *client.lastList.Prefix != "dnd/campaigns/" {
+		t.Fatalf("List() prefix = %#v, want normalized prefix", client.lastList)
+	}
+}
+
+func TestS3BackendDownloadCreatesParentDirectory(t *testing.T) {
+	client := &fakeS3API{getBody: io.NopCloser(strings.NewReader("audio"))}
+	backend := &S3Backend{bucket: "bucket-1", client: client}
+
+	dst := filepath.Join(t.TempDir(), "nested", "clip.flac")
+	if err := backend.Download(context.Background(), `audio\clip.flac`, dst); err != nil {
+		t.Fatalf("Download() error = %v", err)
+	}
+
+	data, err := os.ReadFile(dst)
+	if err != nil {
+		t.Fatalf("ReadFile() error = %v", err)
+	}
+	if string(data) != "audio" {
+		t.Fatalf("downloaded content = %q, want %q", string(data), "audio")
+	}
+	if client.lastGet == nil || *client.lastGet.Key != "audio/clip.flac" {
+		t.Fatalf("GetObject key = %#v, want normalized key", client.lastGet)
+	}
+}
+
+func TestS3BackendReadReturnsOpenedObjectMetadata(t *testing.T) {
+	lastModified := time.Date(2026, 8, 11, 1, 2, 3, 0, time.UTC)
+	client := &fakeS3API{
+		getBody:         io.NopCloser(strings.NewReader("locks")),
+		getSize:         int64Ptr(5),
+		getETag:         strPtr(`"generation"`),
+		getLastModified: &lastModified,
+	}
+	backend := &S3Backend{bucket: "bucket-1", client: client}
+
+	info, body, err := backend.Read(context.Background(), `sessions\locks.yml`)
+	if err != nil {
+		t.Fatalf("Read() error = %v", err)
+	}
+	data, readErr := io.ReadAll(body)
+	closeErr := body.Close()
+	if readErr != nil || closeErr != nil {
+		t.Fatalf("read body error=%v close error=%v", readErr, closeErr)
+	}
+	if info.Key != "sessions/locks.yml" || info.Size != 5 || info.ETag != "generation" || info.LastModified == nil || !info.LastModified.Equal(lastModified) {
+		t.Fatalf("Read() info = %#v, want opened object metadata", info)
+	}
+	if string(data) != "locks" || client.lastGet == nil || *client.lastGet.Key != "sessions/locks.yml" {
+		t.Fatalf("Read() data=%q request=%#v", data, client.lastGet)
+	}
+}
+
+func TestS3BackendUploadAndExists(t *testing.T) {
+	client := &fakeS3API{putOut: &s3.PutObjectOutput{ETag: strPtr(`"etag123"`)}}
+	backend := &S3Backend{bucket: "bucket-1", client: client}
+
+	local := filepath.Join(t.TempDir(), "artifact.txt")
+	if err := os.WriteFile(local, []byte("artifact"), 0o644); err != nil {
+		t.Fatalf("WriteFile() error = %v", err)
+	}
+
+	info, err := backend.Upload(context.Background(), local, `runs\id\artifact.txt`, UploadOptions{
+		Metadata: map[string]string{"kind": "artifact"},
+	})
+	if err != nil {
+		t.Fatalf("Upload() error = %v", err)
+	}
+	if info.Key != "runs/id/artifact.txt" {
+		t.Fatalf("Upload key = %q, want normalized key", info.Key)
+	}
+	if info.ETag != "etag123" {
+		t.Fatalf("Upload ETag = %q, want %q", info.ETag, "etag123")
+	}
+	if client.lastPut == nil || *client.lastPut.Key != "runs/id/artifact.txt" {
+		t.Fatalf("PutObject key = %#v, want normalized key", client.lastPut)
+	}
+
+	ok, err := backend.Exists(context.Background(), "runs/id/artifact.txt")
+	if err != nil {
+		t.Fatalf("Exists() error = %v", err)
+	}
+	if !ok {
+		t.Fatal("Exists() = false, want true")
+	}
+}
+
+func TestS3BackendConditionalUploadUsesProviderPrecondition(t *testing.T) {
+	client := &fakeS3API{putOut: &s3.PutObjectOutput{ETag: strPtr(`"etag123"`)}}
+	backend := &S3Backend{bucket: "bucket-1", client: client}
+	if _, err := backend.UploadConditional(context.Background(), strings.NewReader("payload"), "locks.yml", UploadOptions{}, WriteCondition{MatchETag: "before"}); err != nil {
+		t.Fatalf("UploadConditional() error = %v", err)
+	}
+	if client.lastPut == nil || client.lastPut.IfMatch == nil || *client.lastPut.IfMatch != "before" || client.lastPut.IfNoneMatch != nil {
+		t.Fatalf("PutObject conditional input = %#v", client.lastPut)
+	}
+	client.putErr = &smithy.GenericAPIError{Code: "PreconditionFailed", Message: "changed"}
+	_, err := backend.UploadConditional(context.Background(), strings.NewReader("payload"), "locks.yml", UploadOptions{}, WriteCondition{RequireAbsent: true})
+	if !errors.Is(err, ErrConditionNotMet) {
+		t.Fatalf("UploadConditional() error = %v, want ErrConditionNotMet", err)
+	}
+}
+
+func TestS3BackendUploadMissingLocalFile(t *testing.T) {
+	backend := &S3Backend{bucket: "bucket-1", client: &fakeS3API{}}
+	_, err := backend.Upload(context.Background(), filepath.Join(t.TempDir(), "missing.txt"), "key.txt", UploadOptions{})
+	if err == nil || !strings.Contains(err.Error(), "no such file") {
+		t.Fatalf("Upload() error = %v, want missing local file error", err)
+	}
+}
+
+func TestS3BackendExistsNotFound(t *testing.T) {
+	backend := &S3Backend{
+		bucket: "bucket-1",
+		client: &fakeS3API{
+			headErr: &smithy.GenericAPIError{Code: "NotFound", Message: "missing"},
+		},
+	}
+	ok, err := backend.Exists(context.Background(), "missing-key")
+	if err != nil {
+		t.Fatalf("Exists() error = %v", err)
+	}
+	if ok {
+		t.Fatal("Exists() = true, want false")
+	}
+}
+
+func TestNewS3BackendFromConfigUsesClientOptions(t *testing.T) {
+	original := newS3Client
+	t.Cleanup(func() { newS3Client = original })
+	t.Setenv("OBJECT_STORAGE_KEY_ID", "id-123")
+	t.Setenv("OBJECT_STORAGE_KEY", "secret-abc")
+
+	var got s3ClientOptions
+	newS3Client = func(_ context.Context, opts s3ClientOptions) (s3API, error) {
+		got = opts
+		return &fakeS3API{}, nil
+	}
+
+	backend, err := NewS3BackendFromConfig(context.Background(), config.StorageS3Config{
+		Bucket:         "my-archive",
+		Region:         "us-east-1",
+		Endpoint:       "http://localhost:9000",
+		ForcePathStyle: true,
+	})
+	if err != nil {
+		t.Fatalf("NewS3BackendFromConfig() error = %v", err)
+	}
+	if backend.bucket != "my-archive" {
+		t.Fatalf("backend.bucket = %q, want %q", backend.bucket, "my-archive")
+	}
+	if got.Region != "us-east-1" || got.Endpoint != "http://localhost:9000" || !got.ForcePathStyle {
+		t.Fatalf("client options = %#v, want region/endpoint/path-style values", got)
+	}
+	if got.AccessKeyID != "id-123" || got.SecretKey != "secret-abc" {
+		t.Fatalf("client options credentials = %#v, want env-resolved static credentials", got)
+	}
+}
+
+func TestNewS3BackendFromConfigRequiresBucket(t *testing.T) {
+	_, err := NewS3BackendFromConfig(context.Background(), config.StorageS3Config{})
+	if err == nil || !strings.Contains(err.Error(), "bucket is required") {
+		t.Fatalf("NewS3BackendFromConfig() error = %v, want bucket validation", err)
+	}
+}
+
+func TestNewS3BackendFromConfigFallsBackWhenCredentialEnvMissing(t *testing.T) {
+	original := newS3Client
+	t.Cleanup(func() { newS3Client = original })
+
+	var got s3ClientOptions
+	newS3Client = func(_ context.Context, opts s3ClientOptions) (s3API, error) {
+		got = opts
+		return &fakeS3API{}, nil
+	}
+
+	_, err := NewS3BackendFromConfig(context.Background(), config.StorageS3Config{
+		Bucket:         "my-archive",
+		Region:         "us-east-1",
+		AccessKeyIDEnv: "MISSING_ACCESS_KEY_ID",
+		SecretKeyEnv:   "MISSING_SECRET_KEY",
+	})
+	if err != nil {
+		t.Fatalf("NewS3BackendFromConfig() error = %v", err)
+	}
+	if got.AccessKeyID != "" || got.SecretKey != "" {
+		t.Fatalf("client options credentials = %#v, want empty fallback values", got)
+	}
+}
+
+func strPtr(v string) *string { return &v }
+func int64Ptr(v int64) *int64 { return &v }
+func boolPtr(v bool) *bool    { return &v }
+
+var _ s3API = (*fakeS3API)(nil)

internal/adapters/storage/temp_download.go

@@ -0,0 +1,36 @@
+package storage
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+)
+
+// DownloadObjectToTemp downloads an object into a temporary file and returns
+// the cleaned local path.
+func DownloadObjectToTemp(ctx context.Context, store ObjectStore, key, pattern string) (string, error) {
+	if store == nil {
+		return "", fmt.Errorf("object store is required")
+	}
+	if strings.TrimSpace(pattern) == "" {
+		return "", fmt.Errorf("temp file pattern is required")
+	}
+
+	tmp, err := os.CreateTemp("", pattern)
+	if err != nil {
+		return "", fmt.Errorf("create temp file: %w", err)
+	}
+	path := tmp.Name()
+	if err := tmp.Close(); err != nil {
+		_ = os.Remove(path)
+		return "", fmt.Errorf("close temp file: %w", err)
+	}
+
+	if err := store.Download(ctx, key, path); err != nil {
+		_ = os.Remove(path)
+		return "", err
+	}
+	return filepath.Clean(path), nil
+}

internal/adapters/storage/temp_download_test.go

@@ -0,0 +1,69 @@
+package storage
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func TestDownloadObjectToTempSuccess(t *testing.T) {
+	store := &FakeBackend{}
+	store.SeedObject(FakeObject{Key: "sessions/a/current/run_id.txt", Data: []byte("run-123\n")})
+
+	path, err := DownloadObjectToTemp(context.Background(), store, "sessions/a/current/run_id.txt", "narratio-test-*.txt")
+	if err != nil {
+		t.Fatalf("DownloadObjectToTemp() error = %v", err)
+	}
+	t.Cleanup(func() { _ = os.Remove(path) })
+
+	data, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatalf("ReadFile() error = %v", err)
+	}
+	if string(data) != "run-123\n" {
+		t.Fatalf("downloaded data = %q, want %q", string(data), "run-123\n")
+	}
+}
+
+func TestDownloadObjectToTempFailedDownloadRemovesTempFile(t *testing.T) {
+	sentinel := errors.New("download failed")
+	store := &FakeBackend{DownloadErr: sentinel}
+	pattern := "narratio-test-fail-*.txt"
+	before, err := filepath.Glob(filepath.Join(os.TempDir(), "narratio-test-fail-*.txt"))
+	if err != nil {
+		t.Fatalf("Glob(before) error = %v", err)
+	}
+
+	path, err := DownloadObjectToTemp(context.Background(), store, "sessions/a/current/run_id.txt", pattern)
+	if !errors.Is(err, sentinel) {
+		t.Fatalf("DownloadObjectToTemp() error = %v, want %v", err, sentinel)
+	}
+	if strings.TrimSpace(path) != "" {
+		t.Fatalf("DownloadObjectToTemp() path = %q, want empty on failure", path)
+	}
+	after, err := filepath.Glob(filepath.Join(os.TempDir(), "narratio-test-fail-*.txt"))
+	if err != nil {
+		t.Fatalf("Glob(after) error = %v", err)
+	}
+	if len(after) != len(before) {
+		t.Fatalf("temp file count changed after failed download: before=%d after=%d", len(before), len(after))
+	}
+}
+
+func TestDownloadObjectToTempCallerContextWrappingPreservesCause(t *testing.T) {
+	sentinel := errors.New("object missing")
+	store := &FakeBackend{DownloadErr: sentinel}
+
+	_, err := DownloadObjectToTemp(context.Background(), store, "sessions/a/current/run_id.txt", "narratio-test-*.txt")
+	if err == nil {
+		t.Fatal("DownloadObjectToTemp() error = nil, want error")
+	}
+	err = fmt.Errorf("download run pointer failed: %w", err)
+	if !errors.Is(err, sentinel) {
+		t.Fatalf("wrapped error does not preserve sentinel cause: %v", err)
+	}
+}

internal/adapters/subprocess/diagnostics.go

@@ -0,0 +1,391 @@
+package subprocess
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+	"sync"
+
+	"gitea.maximumdirect.net/eric/narratio/internal/fileops"
+)
+
+const (
+	// MaxStdoutDiagnosticBytes bounds persisted stdout from one external command.
+	MaxStdoutDiagnosticBytes int64 = 8 * 1024 * 1024
+	// MaxStderrDiagnosticBytes bounds persisted stderr from one external command.
+	MaxStderrDiagnosticBytes int64 = 8 * 1024 * 1024
+	diagnosticTailBytes            = 2048
+)
+
+var inheritedEnvironmentNames = map[string]struct{}{
+	"COMSPEC":    {},
+	"HOME":       {},
+	"PATH":       {},
+	"SYSTEMROOT": {},
+	"TMP":        {},
+	"TMPDIR":     {},
+	"TEMP":       {},
+	"WINDIR":     {},
+	// These test-only helper destinations let the adapter package tests exercise
+	// real command invocation without widening the production environment.
+	"AUDITA_HELPER_RECORD_PATH":      {},
+	"AUDITA_HELPER_MODE":             {},
+	"GO_WANT_AUDITA_HELPER":          {},
+	"GO_WANT_SCRIPTORIUM_HELPER":     {},
+	"GO_WANT_SERIATIM_HELPER":        {},
+	"GO_WANT_SUBPROCESS_HELPER":      {},
+	"NOTARIUS_CAPTURE_DIR":           {},
+	"NOTARIUS_RECEIPT_FIXTURE":       {},
+	"SCRIPTORIUM_HELPER_RECORD_PATH": {},
+	"SCRIPTORIUM_HELPER_MODE":        {},
+	"SERIATIM_HELPER_RECORD_PATH":    {},
+	"SERIATIM_HELPER_MODE":           {},
+}
+
+var sensitiveEnvironmentNames = map[string]struct{}{
+	"ANTHROPIC_API_KEY":     {},
+	"API_KEY":               {},
+	"AUDITA_LLM_API_KEY":    {},
+	"AWS_ACCESS_KEY_ID":     {},
+	"AWS_SECRET_ACCESS_KEY": {},
+	"AWS_SESSION_TOKEN":     {},
+	"OPENAI_API_KEY":        {},
+	"OPENROUTER_API_KEY":    {},
+}
+
+type captureLimitError struct {
+	stream string
+	owner  string
+	limit  int64
+}
+
+func (e *captureLimitError) Error() string {
+	return fmt.Sprintf("%s diagnostic capture for %s exceeded %d bytes", e.stream, e.owner, e.limit)
+}
+
+type logWriters struct {
+	files  []*os.File
+	Stdout io.Writer
+	Stderr io.Writer
+
+	limits chan *captureLimitError
+	mu     sync.Mutex
+	limit  *captureLimitError
+	stdout *diagnosticWriter
+	stderr *diagnosticWriter
+}
+
+type diagnosticWriter struct {
+	logs      *logWriters
+	stream    string
+	owner     string
+	target    io.Writer
+	limit     int64
+	received  int64
+	persisted int64
+	redactor  streamRedactor
+	tail      []byte
+}
+
+func openLogWriters(stdoutPath, stderrPath, owner string, sensitiveValues []string) (*logWriters, error) {
+	logs := &logWriters{limits: make(chan *captureLimitError, 1)}
+	cleanStdout := cleanLogPath(stdoutPath)
+	cleanStderr := cleanLogPath(stderrPath)
+
+	stdoutFile, err := openDiagnosticFile(cleanStdout)
+	if err != nil {
+		return nil, fmt.Errorf("open stdout log: %w", err)
+	}
+	stderrFile := stdoutFile
+	if cleanStdout != cleanStderr {
+		stderrFile, err = openDiagnosticFile(cleanStderr)
+		if err != nil {
+			_ = stdoutFile.Close()
+			return nil, fmt.Errorf("open stderr log: %w", err)
+		}
+	}
+	if cleanStdout == cleanStderr {
+		logs.files = []*os.File{stdoutFile}
+	} else {
+		logs.files = []*os.File{stdoutFile, stderrFile}
+	}
+
+	logs.stdout = newDiagnosticWriter(logs, "stdout", owner, stdoutFile, MaxStdoutDiagnosticBytes, sensitiveValues)
+	logs.stderr = newDiagnosticWriter(logs, "stderr", owner, stderrFile, MaxStderrDiagnosticBytes, sensitiveValues)
+	logs.Stdout = logs.stdout
+	logs.Stderr = logs.stderr
+	return logs, nil
+}
+
+func newDiagnosticWriter(logs *logWriters, stream, owner string, target io.Writer, limit int64, sensitiveValues []string) *diagnosticWriter {
+	return &diagnosticWriter{
+		logs:     logs,
+		stream:   stream,
+		owner:    owner,
+		target:   target,
+		limit:    limit,
+		redactor: newStreamRedactor(sensitiveValues),
+	}
+}
+
+func (w *diagnosticWriter) Write(data []byte) (int, error) {
+	if w.received >= w.limit {
+		return len(data), w.reachLimit()
+	}
+	accepted := data
+	if remaining := w.limit - w.received; int64(len(accepted)) > remaining {
+		accepted = accepted[:remaining]
+	}
+	w.received += int64(len(accepted))
+	if err := w.writeRedacted(w.redactor.Write(accepted)); err != nil {
+		return len(data), err
+	}
+	if len(accepted) != len(data) {
+		return len(data), w.reachLimit()
+	}
+	return len(data), nil
+}
+
+func (w *diagnosticWriter) Flush() error {
+	return w.writeRedacted(w.redactor.Flush())
+}
+
+func (w *diagnosticWriter) writeRedacted(data []byte) error {
+	if len(data) == 0 {
+		return nil
+	}
+	w.logs.mu.Lock()
+	remaining := w.limit - w.persisted
+	if remaining <= 0 {
+		w.logs.mu.Unlock()
+		return w.reachLimit()
+	}
+	toWrite := data
+	exceeded := int64(len(data)) > remaining
+	if exceeded {
+		toWrite = toWrite[:remaining]
+	}
+	written, err := w.target.Write(toWrite)
+	w.persisted += int64(written)
+	w.retainTail(toWrite[:written])
+	w.logs.mu.Unlock()
+	if err != nil {
+		return err
+	}
+	if exceeded {
+		return w.reachLimit()
+	}
+	return nil
+}
+
+func (w *diagnosticWriter) Tail() string {
+	w.logs.mu.Lock()
+	defer w.logs.mu.Unlock()
+	return strings.TrimSpace(string(w.tail))
+}
+
+func (w *diagnosticWriter) retainTail(data []byte) {
+	if len(data) >= diagnosticTailBytes {
+		if cap(w.tail) < diagnosticTailBytes {
+			w.tail = make([]byte, diagnosticTailBytes)
+		} else {
+			w.tail = w.tail[:diagnosticTailBytes]
+		}
+		copy(w.tail, data[len(data)-diagnosticTailBytes:])
+		return
+	}
+	if cap(w.tail) < diagnosticTailBytes {
+		retained := make([]byte, len(w.tail), diagnosticTailBytes)
+		copy(retained, w.tail)
+		w.tail = retained
+	}
+	if overflow := len(w.tail) + len(data) - diagnosticTailBytes; overflow > 0 {
+		copy(w.tail, w.tail[overflow:])
+		w.tail = w.tail[:len(w.tail)-overflow]
+	}
+	w.tail = append(w.tail, data...)
+}
+
+func (w *diagnosticWriter) reachLimit() error {
+	limit := &captureLimitError{stream: w.stream, owner: w.owner, limit: w.limit}
+	w.logs.mu.Lock()
+	if w.logs.limit == nil {
+		w.logs.limit = limit
+		w.logs.limits <- limit
+	}
+	w.logs.mu.Unlock()
+	return limit
+}
+
+func (l *logWriters) Limits() <-chan *captureLimitError { return l.limits }
+
+func (l *logWriters) Limit() *captureLimitError {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	return l.limit
+}
+
+func (l *logWriters) Flush() error {
+	return joinErrors(l.stdout.Flush(), l.stderr.Flush())
+}
+
+func (l *logWriters) Close() {
+	_ = l.Flush()
+	for _, file := range l.files {
+		_ = file.Close()
+	}
+}
+
+func cleanLogPath(path string) string {
+	trimmed := strings.TrimSpace(path)
+	if trimmed == "" {
+		return ""
+	}
+	return filepath.Clean(trimmed)
+}
+
+func openDiagnosticFile(path string) (*os.File, error) {
+	if path == "" {
+		return os.OpenFile(os.DevNull, os.O_WRONLY, 0)
+	}
+	if err := fileops.EnsureWorkspaceDirectory(filepath.Dir(path)); err != nil {
+		return nil, fmt.Errorf("create log directory for %q: %w", path, err)
+	}
+	file, err := fileops.OpenFileConfined(path, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, fileops.WorkspaceFileMode)
+	if err != nil {
+		return nil, fmt.Errorf("open log file %q: %w", path, err)
+	}
+	if err := file.Chmod(fileops.WorkspaceFileMode); err != nil {
+		_ = file.Close()
+		return nil, fmt.Errorf("set log file permissions %q: %w", path, err)
+	}
+	return file, nil
+}
+
+func (r RunRequest) diagnosticOwner() string {
+	if owner := strings.TrimSpace(r.DiagnosticOwner); owner != "" {
+		return owner
+	}
+	return "subprocess"
+}
+
+func buildChildEnvironment(base []string, overrides map[string]string) []string {
+	values := make(map[string]string, len(inheritedEnvironmentNames)+len(overrides))
+	for _, item := range base {
+		name, value, ok := strings.Cut(item, "=")
+		if !ok {
+			continue
+		}
+		normalized := strings.ToUpper(name)
+		if _, allowed := inheritedEnvironmentNames[normalized]; allowed {
+			values[name] = value
+		}
+	}
+	for name, value := range overrides {
+		values[name] = value
+	}
+	names := make([]string, 0, len(values))
+	for name := range values {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	out := make([]string, 0, len(names))
+	for _, name := range names {
+		out = append(out, name+"="+values[name])
+	}
+	return out
+}
+
+func sensitiveEnvironmentValues(environment []string, additionalNames []string) []string {
+	names := make(map[string]struct{}, len(sensitiveEnvironmentNames)+len(additionalNames))
+	for name := range sensitiveEnvironmentNames {
+		names[name] = struct{}{}
+	}
+	for _, name := range additionalNames {
+		if trimmed := strings.ToUpper(strings.TrimSpace(name)); trimmed != "" {
+			names[trimmed] = struct{}{}
+		}
+	}
+	values := make([]string, 0, len(names))
+	for _, item := range environment {
+		name, value, ok := strings.Cut(item, "=")
+		if !ok || strings.TrimSpace(value) == "" {
+			continue
+		}
+		if _, sensitive := names[strings.ToUpper(name)]; sensitive {
+			values = append(values, value)
+		}
+	}
+	return values
+}
+
+type streamRedactor struct {
+	values []string
+	buffer []byte
+	maxLen int
+}
+
+func newStreamRedactor(values []string) streamRedactor {
+	unique := make(map[string]struct{}, len(values))
+	for _, value := range values {
+		if value != "" {
+			unique[value] = struct{}{}
+		}
+	}
+	sorted := make([]string, 0, len(unique))
+	for value := range unique {
+		sorted = append(sorted, value)
+	}
+	sort.Slice(sorted, func(i, j int) bool { return len(sorted[i]) > len(sorted[j]) })
+	maxLen := 1
+	for _, value := range sorted {
+		if len(value) > maxLen {
+			maxLen = len(value)
+		}
+	}
+	return streamRedactor{values: sorted, maxLen: maxLen}
+}
+
+func (r *streamRedactor) Write(data []byte) []byte {
+	r.buffer = append(r.buffer, data...)
+	safeCut := len(r.buffer) - r.maxLen + 1
+	if safeCut <= 0 {
+		return nil
+	}
+	emitCut := safeCut
+	for _, value := range r.values {
+		start := 0
+		for {
+			index := bytes.Index(r.buffer[start:], []byte(value))
+			if index < 0 {
+				break
+			}
+			index += start
+			if index+len(value) > safeCut && index < emitCut {
+				emitCut = index
+			}
+			start = index + 1
+		}
+	}
+	output := redactBytes(r.buffer[:emitCut], r.values)
+	r.buffer = append(r.buffer[:0], r.buffer[emitCut:]...)
+	return output
+}
+
+func (r *streamRedactor) Flush() []byte {
+	output := redactBytes(r.buffer, r.values)
+	r.buffer = nil
+	return output
+}
+
+func redactBytes(data []byte, values []string) []byte {
+	out := append([]byte(nil), data...)
+	for _, value := range values {
+		out = bytes.ReplaceAll(out, []byte(value), []byte("<redacted>"))
+	}
+	return out
+}

internal/adapters/subprocess/process_tree.go

@@ -0,0 +1,66 @@
+package subprocess
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os/exec"
+	"time"
+)
+
+const (
+	gracefulTerminationWait = 2 * time.Second
+	forcefulTerminationWait = 2 * time.Second
+)
+
+// ownedProcessTree owns every process started by a command invocation.
+// Implementations must tolerate a leader that has already exited.
+type ownedProcessTree interface {
+	Start(*exec.Cmd) error
+	TerminateGracefully() error
+	TerminateForcefully() error
+	Dispose() error
+}
+
+func waitForOwnedCommand(ctx context.Context, tree ownedProcessTree, waitCh <-chan error, captureLimits <-chan *captureLimitError) (waitErr, ctxErr error, captureLimit *captureLimitError, cleanupErr error) {
+	select {
+	case waitErr = <-waitCh:
+		return waitErr, nil, nil, nil
+	case <-ctx.Done():
+		ctxErr = ctx.Err()
+	case captureLimit = <-captureLimits:
+	}
+
+	cleanupErr = tree.TerminateGracefully()
+	gracefulTimer := time.NewTimer(gracefulTerminationWait)
+	defer gracefulTimer.Stop()
+
+	select {
+	case waitErr = <-waitCh:
+		// The leader may exit before descendants finish graceful shutdown.
+		cleanupErr = joinErrors(cleanupErr, tree.TerminateForcefully())
+		return waitErr, ctxErr, captureLimit, cleanupErr
+	case <-gracefulTimer.C:
+	}
+
+	cleanupErr = joinErrors(cleanupErr, tree.TerminateForcefully())
+	forcefulTimer := time.NewTimer(forcefulTerminationWait)
+	defer forcefulTimer.Stop()
+
+	select {
+	case waitErr = <-waitCh:
+		return waitErr, ctxErr, captureLimit, cleanupErr
+	case <-forcefulTimer.C:
+		return nil, ctxErr, captureLimit, joinErrors(cleanupErr, fmt.Errorf("owned subprocess did not reap within %s after forceful termination", forcefulTerminationWait))
+	}
+}
+
+func joinErrors(errs ...error) error {
+	filtered := make([]error, 0, len(errs))
+	for _, err := range errs {
+		if err != nil {
+			filtered = append(filtered, err)
+		}
+	}
+	return errors.Join(filtered...)
+}

internal/adapters/subprocess/process_tree_supported_test.go

@@ -0,0 +1,227 @@
+//go:build linux || darwin || windows
+
+package subprocess
+
+import (
+	"context"
+	"errors"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+)
+
+func TestRunCancellationTerminatesProcessTree(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	resultCh := make(chan runOutcome, 1)
+	req, sentinelPath := processTreeRequest(t)
+	go func() {
+		result, err := Run(ctx, req)
+		resultCh <- runOutcome{result: result, err: err}
+	}()
+
+	awaitHelperReady(t, req.EnvOverrides["SUBPROCESS_HELPER_READY_PATH"])
+	cancel()
+
+	outcome := awaitRunOutcome(t, resultCh)
+	if !outcome.result.Canceled {
+		t.Fatalf("Canceled = %v, want true", outcome.result.Canceled)
+	}
+	if !errors.Is(outcome.err, context.Canceled) {
+		t.Fatalf("error = %v, want context cancellation", outcome.err)
+	}
+	assertDescendantDidNotSurvive(t, sentinelPath)
+}
+
+func TestRunTimeoutTerminatesProcessTree(t *testing.T) {
+	req, sentinelPath := processTreeRequest(t)
+	req.Timeout = 100 * time.Millisecond
+
+	result, err := Run(context.Background(), req)
+	if !result.TimedOut {
+		t.Fatalf("TimedOut = %v, want true", result.TimedOut)
+	}
+	if !errors.Is(err, context.DeadlineExceeded) {
+		t.Fatalf("error = %v, want context deadline exceeded", err)
+	}
+	assertDescendantDidNotSurvive(t, sentinelPath)
+}
+
+func TestRunCaptureLimitTerminatesProcessTree(t *testing.T) {
+	req, sentinelPath := processTreeRequest(t)
+	req.Args[len(req.Args)-1] = "tree-spam"
+
+	result, err := Run(context.Background(), req)
+	if err == nil {
+		t.Fatal("Run() error = nil, want capture-limit error")
+	}
+	if result.ExitCode == 0 {
+		t.Fatalf("ExitCode = %d, want terminated process", result.ExitCode)
+	}
+	if !strings.Contains(err.Error(), "stdout diagnostic capture for subprocess exceeded") {
+		t.Fatalf("error = %q, want stdout capture-limit context", err)
+	}
+	info, statErr := os.Stat(req.StdoutLogPath)
+	if statErr != nil {
+		t.Fatalf("stat stdout diagnostic: %v", statErr)
+	}
+	if info.Size() != MaxStdoutDiagnosticBytes {
+		t.Fatalf("stdout diagnostic size = %d, want %d", info.Size(), MaxStdoutDiagnosticBytes)
+	}
+	assertDescendantDidNotSurvive(t, sentinelPath)
+}
+
+func TestRunDisposesDescendantsAfterLeaderExit(t *testing.T) {
+	tests := []struct {
+		name          string
+		mode          string
+		wantExitCode  int
+		wantWaitDelay bool
+		ignoreTerm    bool
+	}{
+		{name: "success retaining streams", mode: "leader-exit-retained", wantExitCode: 0, wantWaitDelay: true},
+		{name: "success redirecting streams", mode: "leader-exit-redirected", wantExitCode: 0},
+		{name: "failed leader", mode: "leader-fail-redirected", wantExitCode: 9, ignoreTerm: true},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			req, sentinelPath, releasePath := leaderExitRequest(t, tt.mode)
+			if tt.ignoreTerm {
+				req.EnvOverrides["SUBPROCESS_HELPER_IGNORE_TERM"] = "1"
+			}
+			outcomes := make(chan runOutcome, 1)
+			go func() {
+				result, err := Run(context.Background(), req)
+				outcomes <- runOutcome{result: result, err: err}
+			}()
+
+			var outcome runOutcome
+			select {
+			case outcome = <-outcomes:
+			case <-time.After(6 * time.Second):
+				t.Fatal("Run() did not complete bounded owned-tree disposal")
+			}
+			if outcome.result.ExitCode != tt.wantExitCode {
+				t.Fatalf("ExitCode = %d, want %d", outcome.result.ExitCode, tt.wantExitCode)
+			}
+			if tt.wantWaitDelay {
+				if !errors.Is(outcome.err, exec.ErrWaitDelay) {
+					t.Fatalf("error = %v, want exec.ErrWaitDelay", outcome.err)
+				}
+			} else if tt.wantExitCode == 0 && outcome.err != nil {
+				t.Fatalf("Run() error = %v, want nil", outcome.err)
+			} else if tt.wantExitCode != 0 {
+				var exitErr *exec.ExitError
+				if !errors.As(outcome.err, &exitErr) || exitErr.ExitCode() != tt.wantExitCode {
+					t.Fatalf("error = %v, want exit code %d", outcome.err, tt.wantExitCode)
+				}
+			}
+			if _, err := os.Stat(req.EnvOverrides["SUBPROCESS_HELPER_READY_PATH"]); err != nil {
+				t.Fatalf("descendant readiness file: %v", err)
+			}
+			if err := os.WriteFile(releasePath, []byte("release"), 0o600); err != nil {
+				t.Fatalf("WriteFile(release) error = %v", err)
+			}
+			assertDescendantDidNotSurvive(t, sentinelPath)
+		})
+	}
+}
+
+type runOutcome struct {
+	result RunResult
+	err    error
+}
+
+func processTreeRequest(t *testing.T) (RunRequest, string) {
+	t.Helper()
+
+	executable, err := os.Executable()
+	if err != nil {
+		t.Fatalf("os.Executable() error = %v", err)
+	}
+	dir := t.TempDir()
+	readyPath := filepath.Join(dir, "ready")
+	sentinelPath := filepath.Join(dir, "descendant-survived")
+	return RunRequest{
+		Executable: executable,
+		Args:       []string{"-test.run=^TestSubprocessHelper$", "--", "tree"},
+		EnvOverrides: map[string]string{
+			"GO_WANT_SUBPROCESS_HELPER":       "1",
+			"SUBPROCESS_HELPER_READY_PATH":    readyPath,
+			"SUBPROCESS_HELPER_SENTINEL_PATH": sentinelPath,
+		},
+		StdoutLogPath: filepath.Join(dir, "stdout.log"),
+		StderrLogPath: filepath.Join(dir, "stderr.log"),
+	}, sentinelPath
+}
+
+func leaderExitRequest(t *testing.T, mode string) (RunRequest, string, string) {
+	t.Helper()
+
+	executable, err := os.Executable()
+	if err != nil {
+		t.Fatalf("os.Executable() error = %v", err)
+	}
+	dir := t.TempDir()
+	readyPath := filepath.Join(dir, "ready")
+	releasePath := filepath.Join(dir, "release")
+	sentinelPath := filepath.Join(dir, "descendant-survived")
+	return RunRequest{
+		Executable: executable,
+		Args:       []string{"-test.run=^TestSubprocessHelper$", "--", mode},
+		EnvOverrides: map[string]string{
+			"GO_WANT_SUBPROCESS_HELPER":       "1",
+			"SUBPROCESS_HELPER_READY_PATH":    readyPath,
+			"SUBPROCESS_HELPER_RELEASE_PATH":  releasePath,
+			"SUBPROCESS_HELPER_SENTINEL_PATH": sentinelPath,
+		},
+		StdoutLogPath: filepath.Join(dir, "stdout.log"),
+		StderrLogPath: filepath.Join(dir, "stderr.log"),
+	}, sentinelPath, releasePath
+}
+
+func awaitHelperReady(t *testing.T, readyPath string) {
+	t.Helper()
+
+	deadline := time.Now().Add(2 * time.Second)
+	for time.Now().Before(deadline) {
+		if _, err := os.Stat(readyPath); err == nil {
+			return
+		} else if !errors.Is(err, os.ErrNotExist) {
+			t.Fatalf("stat helper readiness: %v", err)
+		}
+		time.Sleep(10 * time.Millisecond)
+	}
+	t.Fatal("helper did not start its descendant")
+}
+
+func awaitRunOutcome(t *testing.T, outcomes <-chan runOutcome) runOutcome {
+	t.Helper()
+
+	select {
+	case outcome := <-outcomes:
+		if outcome.err == nil {
+			t.Fatal("Run() error = nil, want cancellation error")
+		}
+		return outcome
+	case <-time.After(3 * time.Second):
+		t.Fatal("Run() did not return after cancellation")
+		return runOutcome{}
+	}
+}
+
+func assertDescendantDidNotSurvive(t *testing.T, sentinelPath string) {
+	t.Helper()
+
+	time.Sleep(700 * time.Millisecond)
+	if _, err := os.Stat(sentinelPath); err == nil {
+		t.Fatal("descendant survived cancellation and wrote its sentinel")
+	} else if !errors.Is(err, os.ErrNotExist) {
+		t.Fatalf("stat descendant sentinel: %v", err)
+	}
+}